modelId
string | author
string | last_modified
timestamp[us, tz=UTC] | downloads
int64 | likes
int64 | library_name
string | tags
sequence | pipeline_tag
string | createdAt
timestamp[us, tz=UTC] | card
string |
|---|---|---|---|---|---|---|---|---|---|
Mungert/phi-4-GGUF | Mungert | 2025-06-15T19:41:59Z | 2,234 | 5 | transformers | [
"transformers",
"gguf",
"phi",
"nlp",
"math",
"code",
"chat",
"conversational",
"text-generation",
"en",
"arxiv:2412.08905",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix"
] | text-generation | 2025-03-23T09:15:42Z | ---
license: mit
license_link: https://huggingface.co/microsoft/phi-4/resolve/main/LICENSE
language:
- en
pipeline_tag: text-generation
tags:
- phi
- nlp
- math
- code
- chat
- conversational
inference:
parameters:
temperature: 0
widget:
- messages:
- role: user
content: How should I explain the Internet?
library_name: transformers
---
# <span style="color: #7FFF7F;">phi-4 GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `phi-4-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `phi-4-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `phi-4-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `phi-4-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `phi-4-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `phi-4-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `phi-4-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `phi-4-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `phi-4-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `phi-4-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `phi-4-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Phi-4 Model Card
[Phi-4 Technical Report](https://arxiv.org/pdf/2412.08905)
## Model Summary
| | |
|-------------------------|-------------------------------------------------------------------------------|
| **Developers** | Microsoft Research |
| **Description** | `phi-4` is a state-of-the-art open model built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets. The goal of this approach was to ensure that small capable models were trained with data focused on high quality and advanced reasoning.<br><br>`phi-4` underwent a rigorous enhancement and alignment process, incorporating both supervised fine-tuning and direct preference optimization to ensure precise instruction adherence and robust safety measures |
| **Architecture** | 14B parameters, dense decoder-only Transformer model |
| **Inputs** | Text, best suited for prompts in the chat format |
| **Context length** | 16K tokens |
| **GPUs** | 1920 H100-80G |
| **Training time** | 21 days |
| **Training data** | 9.8T tokens |
| **Outputs** | Generated text in response to input |
| **Dates** | October 2024 – November 2024 |
| **Status** | Static model trained on an offline dataset with cutoff dates of June 2024 and earlier for publicly available data |
| **Release date** | December 12, 2024 |
| **License** | MIT |
## Intended Use
| | |
|-------------------------------|-------------------------------------------------------------------------|
| **Primary Use Cases** | Our model is designed to accelerate research on language models, for use as a building block for generative AI powered features. It provides uses for general purpose AI systems and applications (primarily in English) which require:<br><br>1. Memory/compute constrained environments.<br>2. Latency bound scenarios.<br>3. Reasoning and logic. |
| **Out-of-Scope Use Cases** | Our models is not specifically designed or evaluated for all downstream purposes, thus:<br><br>1. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios.<br>2. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case, including the model’s focus on English.<br>3. Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under. |
## Data Overview
### Training Datasets
Our training data is an extension of the data used for Phi-3 and includes a wide variety of sources from:
1. Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code.
2. Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.).
3. Acquired academic books and Q&A datasets.
4. High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.
Multilingual data constitutes about 8% of our overall data. We are focusing on the quality of data that could potentially improve the reasoning ability for the model, and we filter the publicly available documents to contain the correct level of knowledge.
#### Benchmark datasets
We evaluated `phi-4` using [OpenAI’s SimpleEval](https://github.com/openai/simple-evals) and our own internal benchmarks to understand the model’s capabilities, more specifically:
* **MMLU:** Popular aggregated dataset for multitask language understanding.
* **MATH:** Challenging competition math problems.
* **GPQA:** Complex, graduate-level science questions.
* **DROP:** Complex comprehension and reasoning.
* **MGSM:** Multi-lingual grade-school math.
* **HumanEval:** Functional code generation.
* **SimpleQA:** Factual responses.
## Safety
### Approach
`phi-4` has adopted a robust safety post-training approach. This approach leverages a variety of both open-source and in-house generated synthetic datasets. The overall technique employed to do the safety alignment is a combination of SFT (Supervised Fine-Tuning) and iterative DPO (Direct Preference Optimization), including publicly available datasets focusing on helpfulness and harmlessness as well as various questions and answers targeted to multiple safety categories.
### Safety Evaluation and Red-Teaming
Prior to release, `phi-4` followed a multi-faceted evaluation approach. Quantitative evaluation was conducted with multiple open-source safety benchmarks and in-house tools utilizing adversarial conversation simulation. For qualitative safety evaluation, we collaborated with the independent AI Red Team (AIRT) at Microsoft to assess safety risks posed by `phi-4` in both average and adversarial user scenarios. In the average user scenario, AIRT emulated typical single-turn and multi-turn interactions to identify potentially risky behaviors. The adversarial user scenario tested a wide range of techniques aimed at intentionally subverting the model’s safety training including jailbreaks, encoding-based attacks, multi-turn attacks, and adversarial suffix attacks.
Please refer to the technical report for more details on safety alignment.
## Model Quality
To understand the capabilities, we compare `phi-4` with a set of models over OpenAI’s SimpleEval benchmark.
At the high-level overview of the model quality on representative benchmarks. For the table below, higher numbers indicate better performance:
| **Category** | **Benchmark** | **phi-4** (14B) | **phi-3** (14B) | **Qwen 2.5** (14B instruct) | **GPT-4o-mini** | **Llama-3.3** (70B instruct) | **Qwen 2.5** (72B instruct) | **GPT-4o** |
|------------------------------|---------------|-----------|-----------------|----------------------|----------------------|--------------------|-------------------|-----------------|
| Popular Aggregated Benchmark | MMLU | 84.8 | 77.9 | 79.9 | 81.8 | 86.3 | 85.3 | **88.1** |
| Science | GPQA | **56.1** | 31.2 | 42.9 | 40.9 | 49.1 | 49.0 | 50.6 |
| Math | MGSM<br>MATH | 80.6<br>**80.4** | 53.5<br>44.6 | 79.6<br>75.6 | 86.5<br>73.0 | 89.1<br>66.3* | 87.3<br>80.0 | **90.4**<br>74.6 |
| Code Generation | HumanEval | 82.6 | 67.8 | 72.1 | 86.2 | 78.9* | 80.4 | **90.6** |
| Factual Knowledge | SimpleQA | 3.0 | 7.6 | 5.4 | 9.9 | 20.9 | 10.2 | **39.4** |
| Reasoning | DROP | 75.5 | 68.3 | 85.5 | 79.3 | **90.2** | 76.7 | 80.9 |
\* These scores are lower than those reported by Meta, perhaps because simple-evals has a strict formatting requirement that Llama models have particular trouble following. We use the simple-evals framework because it is reproducible, but Meta reports 77 for MATH and 88 for HumanEval on Llama-3.3-70B.
## Usage
### Input Formats
Given the nature of the training data, `phi-4` is best suited for prompts using the chat format as follows:
```bash
<|im_start|>system<|im_sep|>
You are a medieval knight and must provide explanations to modern people.<|im_end|>
<|im_start|>user<|im_sep|>
How should I explain the Internet?<|im_end|>
<|im_start|>assistant<|im_sep|>
```
### With `transformers`
```python
import transformers
pipeline = transformers.pipeline(
"text-generation",
model="microsoft/phi-4",
model_kwargs={"torch_dtype": "auto"},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a medieval knight and must provide explanations to modern people."},
{"role": "user", "content": "How should I explain the Internet?"},
]
outputs = pipeline(messages, max_new_tokens=128)
print(outputs[0]["generated_text"][-1])
```
## Responsible AI Considerations
Like other language models, `phi-4` can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
* **Quality of Service:** The model is trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English. `phi-4` is not intended to support multilingual use.
* **Representation of Harms & Perpetuation of Stereotypes:** These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
* **Inappropriate or Offensive Content:** These models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case.
* **Information Reliability:** Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
* **Limited Scope for Code:** Majority of `phi-4` training data is based in Python and uses common packages such as `typing`, `math`, `random`, `collections`, `datetime`, `itertools`. If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.
Developers should apply responsible AI best practices and are responsible for ensuring that a specific use case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Using safety services like [Azure AI Content Safety](https://azure.microsoft.com/en-us/products/ai-services/ai-content-safety) that have advanced guardrails is highly recommended. Important areas for consideration include:
* **Allocation:** Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
* **High-Risk Scenarios:** Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
* **Misinformation:** Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
* **Generation of Harmful Content:** Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
* **Misuse:** Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations. |
Mungert/Llama-3.1-8B-GGUF | Mungert | 2025-06-15T19:41:50Z | 255 | 0 | transformers | [
"transformers",
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"de",
"fr",
"it",
"pt",
"hi",
"es",
"th",
"arxiv:2204.05149",
"license:llama3.1",
"endpoints_compatible",
"region:us",
"imatrix"
] | text-generation | 2025-03-22T20:20:18Z | ---
language:
- en
- de
- fr
- it
- pt
- hi
- es
- th
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3.1
extra_gated_prompt: >-
### LLAMA 3.1 COMMUNITY LICENSE AGREEMENT
Llama 3.1 Version Release Date: July 23, 2024
"Agreement" means the terms and conditions for use, reproduction, distribution
and modification of the Llama Materials set forth herein.
"Documentation" means the specifications, manuals and documentation
accompanying Llama 3.1 distributed by Meta at
https://llama.meta.com/doc/overview.
"Licensee" or "you" means you, or your employer or any other person or entity
(if you are entering into this Agreement on such person or entity’s behalf),
of the age required under applicable laws, rules or regulations to provide
legal consent and that has legal authority to bind your employer or such other
person or entity if you are entering in this Agreement on their behalf.
"Llama 3.1" means the foundational large language models and software and
algorithms, including machine-learning model code, trained model weights,
inference-enabling code, training-enabling code, fine-tuning enabling code and
other elements of the foregoing distributed by Meta at
https://llama.meta.com/llama-downloads.
"Llama Materials" means, collectively, Meta’s proprietary Llama 3.1 and
Documentation (and any portion thereof) made available under this Agreement.
"Meta" or "we" means Meta Platforms Ireland Limited (if you are located in or,
if you are an entity, your principal place of business is in the EEA or
Switzerland) and Meta Platforms, Inc. (if you are located outside of the EEA
or Switzerland).
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide,
non-transferable and royalty-free limited license under Meta’s intellectual
property or other rights owned by Meta embodied in the Llama Materials to use,
reproduce, distribute, copy, create derivative works of, and make
modifications to the Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative
works thereof), or a product or service (including another AI model) that
contains any of them, you shall (A) provide a copy of this Agreement with any
such Llama Materials; and (B) prominently display “Built with Llama” on a
related website, user interface, blogpost, about page, or product
documentation. If you use the Llama Materials or any outputs or results of the
Llama Materials to create, train, fine tune, or otherwise improve an AI model,
which is distributed or made available, you shall also include “Llama” at the
beginning of any such AI model name.
ii. If you receive Llama Materials, or any derivative works thereof, from a
Licensee as part of an integrated end user product, then Section 2 of this
Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute
the following attribution notice within a “Notice” text file distributed as a
part of such copies: “Llama 3.1 is licensed under the Llama 3.1 Community
License, Copyright © Meta Platforms, Inc. All Rights Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and
regulations (including trade compliance laws and regulations) and adhere to
the Acceptable Use Policy for the Llama Materials (available at
https://llama.meta.com/llama3_1/use-policy), which is hereby incorporated by
reference into this Agreement.
2. Additional Commercial Terms. If, on the Llama 3.1 version release date, the
monthly active users of the products or services made available by or for
Licensee, or Licensee’s affiliates, is greater than 700 million monthly active
users in the preceding calendar month, you must request a license from Meta,
which Meta may grant to you in its sole discretion, and you are not authorized
to exercise any of the rights under this Agreement unless or until Meta
otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA
MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS”
BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF
ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY
WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A
PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE
APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY
RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE
UNDER ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS
LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS
OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE
DAMAGES, EVEN IF META OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY
OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection
with the Llama Materials, neither Meta nor Licensee may use any name or mark
owned by or associated with the other or any of its affiliates, except as
required for reasonable and customary use in describing and redistributing the
Llama Materials or as set forth in this Section 5(a). Meta hereby grants you a
license to use “Llama” (the “Mark”) solely as required to comply with the last
sentence of Section 1.b.i. You will comply with Meta’s brand guidelines
(currently accessible at
https://about.meta.com/brand/resources/meta/company-brand/ ). All goodwill
arising out of your use of the Mark will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or
for Meta, with respect to any derivative works and modifications of the Llama
Materials that are made by you, as between you and Meta, you are and will be
the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity
(including a cross-claim or counterclaim in a lawsuit) alleging that the Llama
Materials or Llama 3.1 outputs or results, or any portion of any of the
foregoing, constitutes infringement of intellectual property or other rights
owned or licensable by you, then any licenses granted to you under this
Agreement shall terminate as of the date such litigation or claim is filed or
instituted. You will indemnify and hold harmless Meta from and against any
claim by any third party arising out of or related to your use or distribution
of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your
acceptance of this Agreement or access to the Llama Materials and will
continue in full force and effect until terminated in accordance with the
terms and conditions herein. Meta may terminate this Agreement if you are in
breach of any term or condition of this Agreement. Upon termination of this
Agreement, you shall delete and cease use of the Llama Materials. Sections 3,
4 and 7 shall survive the termination of this Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and
construed under the laws of the State of California without regard to choice
of law principles, and the UN Convention on Contracts for the International
Sale of Goods does not apply to this Agreement. The courts of California shall
have exclusive jurisdiction of any dispute arising out of this Agreement.
### Llama 3.1 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features,
including Llama 3.1. If you access or use Llama 3.1, you agree to this
Acceptable Use Policy (“Policy”). The most recent copy of this policy can be
found at
[https://llama.meta.com/llama3_1/use-policy](https://llama.meta.com/llama3_1/use-policy)
#### Prohibited Uses
We want everyone to use Llama 3.1 safely and responsibly. You agree you will
not use, or allow others to use, Llama 3.1 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
3. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
4. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
5. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
6. Collect, process, disclose, generate, or infer health, demographic, or other sensitive personal or private information about individuals without rights and consents required by applicable laws
7. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
8. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
2. Engage in, promote, incite, facilitate, or assist in the planning or
development of activities that present a risk of death or bodily harm to
individuals, including use of Llama 3.1 related to the following:
1. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State
2. Guns and illegal weapons (including weapon development)
3. Illegal drugs and regulated/controlled substances
4. Operation of critical infrastructure, transportation technologies, or heavy machinery
5. Self-harm or harm to others, including suicide, cutting, and eating disorders
6. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Llama 3.1 related
to the following:
1. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
2. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
3. Generating, promoting, or further distributing spam
4. Impersonating another individual without consent, authorization, or legal right
5. Representing that the use of Llama 3.1 or outputs are human-generated
6. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI
system
Please report any violation of this Policy, software “bug,” or other problems
that could lead to a violation of this Policy through one of the following
means:
* Reporting issues with the model: [https://github.com/meta-llama/llama-models/issues](https://github.com/meta-llama/llama-models/issues)
* Reporting risky content generated by the model:
developers.facebook.com/llama_output_feedback
* Reporting bugs and security concerns: facebook.com/whitehat/info
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Meta Llama 3: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Job title:
type: select
options:
- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: >-
The information you provide will be collected, stored, processed and shared in
accordance with the [Meta Privacy
Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
library_name: transformers
---
# <span style="color: #7FFF7F;">Llama-3.1-8B GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.1-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.1-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.1-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.1-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.1-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.1-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.1-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.1-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.1-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.1-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.1-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
## Model Information
The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models in 8B, 70B and 405B sizes (text in/text out). The Llama 3.1 instruction tuned text only models (8B, 70B, 405B) are optimized for multilingual dialogue use cases and outperform many of the available open source and closed chat models on common industry benchmarks.
**Model developer**: Meta
**Model Architecture:** Llama 3.1 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
<table>
<tr>
<td>
</td>
<td><strong>Training Data</strong>
</td>
<td><strong>Params</strong>
</td>
<td><strong>Input modalities</strong>
</td>
<td><strong>Output modalities</strong>
</td>
<td><strong>Context length</strong>
</td>
<td><strong>GQA</strong>
</td>
<td><strong>Token count</strong>
</td>
<td><strong>Knowledge cutoff</strong>
</td>
</tr>
<tr>
<td rowspan="3" >Llama 3.1 (text only)
</td>
<td rowspan="3" >A new mix of publicly available online data.
</td>
<td>8B
</td>
<td>Multilingual Text
</td>
<td>Multilingual Text and code
</td>
<td>128k
</td>
<td>Yes
</td>
<td rowspan="3" >15T+
</td>
<td rowspan="3" >December 2023
</td>
</tr>
<tr>
<td>70B
</td>
<td>Multilingual Text
</td>
<td>Multilingual Text and code
</td>
<td>128k
</td>
<td>Yes
</td>
</tr>
<tr>
<td>405B
</td>
<td>Multilingual Text
</td>
<td>Multilingual Text and code
</td>
<td>128k
</td>
<td>Yes
</td>
</tr>
</table>
**Supported languages:** English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
**Llama 3.1 family of models**. Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date:** July 23, 2024.
**Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.
**License:** A custom commercial license, the Llama 3.1 Community License, is available at: [https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE)
Where to send questions or comments about the model Instructions on how to provide feedback or comments on the model can be found in the model [README](https://github.com/meta-llama/llama3). For more technical information about generation parameters and recipes for how to use Llama 3.1 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases** Llama 3.1 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks. The Llama 3.1 model collection also supports the ability to leverage the outputs of its models to improve other models including synthetic data generation and distillation. The Llama 3.1 Community License allows for these use cases.
**Out-of-scope** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3.1 Community License. Use in languages beyond those explicitly referenced as supported in this model card**.
**<span style="text-decoration:underline;">Note</span>: Llama 3.1 has been trained on a broader collection of languages than the 8 supported languages. Developers may fine-tune Llama 3.1 models for languages beyond the 8 supported languages provided they comply with the Llama 3.1 Community License and the Acceptable Use Policy and in such cases are responsible for ensuring that any uses of Llama 3.1 in additional languages is done in a safe and responsible manner.
## How to use
This repository contains two versions of Meta's Llama-3.1-8B, for use with transformers and with the original `llama` codebase.
### Use with transformers
Starting with transformers >= 4.43.0 onward, you can run conversational inference using the Transformers pipeline abstraction or by leveraging the Auto classes with the generate() function.
Make sure to update your transformers installation via pip install --upgrade transformers.
```python
import transformers
import torch
model_id = "meta-llama/Llama-3.1-8B"
pipeline = transformers.pipeline(
"text-generation", model=model_id, model_kwargs={"torch_dtype": torch.bfloat16}, device_map="auto"
)
pipeline("Hey how are you doing today?")
```
### Use with `llama`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama).
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Llama-3.1-8B --include "original/*" --local-dir Llama-3.1-8B
```
## Hardware and Software
**Training Factors** We used custom training libraries, Meta's custom built GPU cluster, and production infrastructure for pretraining. Fine-tuning, annotation, and evaluation were also performed on production infrastructure.
**Training utilized a cumulative of** 39.3M GPU hours of computation on H100-80GB (TDP of 700W) type hardware, per the table below. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
**Training Greenhouse Gas Emissions** Estimated total location-based greenhouse gas emissions were **11,390** tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with renewable energy, therefore the total market-based greenhouse gas emissions for training were 0 tons CO2eq.
<table>
<tr>
<td>
</td>
<td><strong>Training Time (GPU hours)</strong>
</td>
<td><strong>Training Power Consumption (W)</strong>
</td>
<td><strong>Training Location-Based Greenhouse Gas Emissions</strong>
<p>
<strong>(tons CO2eq)</strong>
</td>
<td><strong>Training Market-Based Greenhouse Gas Emissions</strong>
<p>
<strong>(tons CO2eq)</strong>
</td>
</tr>
<tr>
<td>Llama 3.1 8B
</td>
<td>1.46M
</td>
<td>700
</td>
<td>420
</td>
<td>0
</td>
</tr>
<tr>
<td>Llama 3.1 70B
</td>
<td>7.0M
</td>
<td>700
</td>
<td>2,040
</td>
<td>0
</td>
</tr>
<tr>
<td>Llama 3.1 405B
</td>
<td>30.84M
</td>
<td>700
</td>
<td>8,930
</td>
<td>0
</td>
</tr>
<tr>
<td>Total
</td>
<td>39.3M
<td>
<ul>
</ul>
</td>
<td>11,390
</td>
<td>0
</td>
</tr>
</table>
The methodology used to determine training energy use and greenhouse gas emissions can be found [here](https://arxiv.org/pdf/2204.05149). Since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others.
## Training Data
**Overview:** Llama 3.1 was pretrained on ~15 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over 25M synthetically generated examples.
**Data Freshness:** The pretraining data has a cutoff of December 2023.
## Benchmark scores
In this section, we report the results for Llama 3.1 models on standard automatic benchmarks. For all the evaluations, we use our internal evaluations library.
### Base pretrained models
<table>
<tr>
<td><strong>Category</strong>
</td>
<td><strong>Benchmark</strong>
</td>
<td><strong># Shots</strong>
</td>
<td><strong>Metric</strong>
</td>
<td><strong>Llama 3 8B</strong>
</td>
<td><strong>Llama 3.1 8B</strong>
</td>
<td><strong>Llama 3 70B</strong>
</td>
<td><strong>Llama 3.1 70B</strong>
</td>
<td><strong>Llama 3.1 405B</strong>
</td>
</tr>
<tr>
<td rowspan="7" >General
</td>
<td>MMLU
</td>
<td>5
</td>
<td>macro_avg/acc_char
</td>
<td>66.7
</td>
<td>66.7
</td>
<td>79.5
</td>
<td>79.3
</td>
<td>85.2
</td>
</tr>
<tr>
<td>MMLU-Pro (CoT)
</td>
<td>5
</td>
<td>macro_avg/acc_char
</td>
<td>36.2
</td>
<td>37.1
</td>
<td>55.0
</td>
<td>53.8
</td>
<td>61.6
</td>
</tr>
<tr>
<td>AGIEval English
</td>
<td>3-5
</td>
<td>average/acc_char
</td>
<td>47.1
</td>
<td>47.8
</td>
<td>63.0
</td>
<td>64.6
</td>
<td>71.6
</td>
</tr>
<tr>
<td>CommonSenseQA
</td>
<td>7
</td>
<td>acc_char
</td>
<td>72.6
</td>
<td>75.0
</td>
<td>83.8
</td>
<td>84.1
</td>
<td>85.8
</td>
</tr>
<tr>
<td>Winogrande
</td>
<td>5
</td>
<td>acc_char
</td>
<td>-
</td>
<td>60.5
</td>
<td>-
</td>
<td>83.3
</td>
<td>86.7
</td>
</tr>
<tr>
<td>BIG-Bench Hard (CoT)
</td>
<td>3
</td>
<td>average/em
</td>
<td>61.1
</td>
<td>64.2
</td>
<td>81.3
</td>
<td>81.6
</td>
<td>85.9
</td>
</tr>
<tr>
<td>ARC-Challenge
</td>
<td>25
</td>
<td>acc_char
</td>
<td>79.4
</td>
<td>79.7
</td>
<td>93.1
</td>
<td>92.9
</td>
<td>96.1
</td>
</tr>
<tr>
<td>Knowledge reasoning
</td>
<td>TriviaQA-Wiki
</td>
<td>5
</td>
<td>em
</td>
<td>78.5
</td>
<td>77.6
</td>
<td>89.7
</td>
<td>89.8
</td>
<td>91.8
</td>
</tr>
<tr>
<td rowspan="4" >Reading comprehension
</td>
<td>SQuAD
</td>
<td>1
</td>
<td>em
</td>
<td>76.4
</td>
<td>77.0
</td>
<td>85.6
</td>
<td>81.8
</td>
<td>89.3
</td>
</tr>
<tr>
<td>QuAC (F1)
</td>
<td>1
</td>
<td>f1
</td>
<td>44.4
</td>
<td>44.9
</td>
<td>51.1
</td>
<td>51.1
</td>
<td>53.6
</td>
</tr>
<tr>
<td>BoolQ
</td>
<td>0
</td>
<td>acc_char
</td>
<td>75.7
</td>
<td>75.0
</td>
<td>79.0
</td>
<td>79.4
</td>
<td>80.0
</td>
</tr>
<tr>
<td>DROP (F1)
</td>
<td>3
</td>
<td>f1
</td>
<td>58.4
</td>
<td>59.5
</td>
<td>79.7
</td>
<td>79.6
</td>
<td>84.8
</td>
</tr>
</table>
### Instruction tuned models
<table>
<tr>
<td><strong>Category</strong>
</td>
<td><strong>Benchmark</strong>
</td>
<td><strong># Shots</strong>
</td>
<td><strong>Metric</strong>
</td>
<td><strong>Llama 3 8B Instruct</strong>
</td>
<td><strong>Llama 3.1 8B Instruct</strong>
</td>
<td><strong>Llama 3 70B Instruct</strong>
</td>
<td><strong>Llama 3.1 70B Instruct</strong>
</td>
<td><strong>Llama 3.1 405B Instruct</strong>
</td>
</tr>
<tr>
<td rowspan="4" >General
</td>
<td>MMLU
</td>
<td>5
</td>
<td>macro_avg/acc
</td>
<td>68.5
</td>
<td>69.4
</td>
<td>82.0
</td>
<td>83.6
</td>
<td>87.3
</td>
</tr>
<tr>
<td>MMLU (CoT)
</td>
<td>0
</td>
<td>macro_avg/acc
</td>
<td>65.3
</td>
<td>73.0
</td>
<td>80.9
</td>
<td>86.0
</td>
<td>88.6
</td>
</tr>
<tr>
<td>MMLU-Pro (CoT)
</td>
<td>5
</td>
<td>micro_avg/acc_char
</td>
<td>45.5
</td>
<td>48.3
</td>
<td>63.4
</td>
<td>66.4
</td>
<td>73.3
</td>
</tr>
<tr>
<td>IFEval
</td>
<td>
</td>
<td>
</td>
<td>76.8
</td>
<td>80.4
</td>
<td>82.9
</td>
<td>87.5
</td>
<td>88.6
</td>
</tr>
<tr>
<td rowspan="2" >Reasoning
</td>
<td>ARC-C
</td>
<td>0
</td>
<td>acc
</td>
<td>82.4
</td>
<td>83.4
</td>
<td>94.4
</td>
<td>94.8
</td>
<td>96.9
</td>
</tr>
<tr>
<td>GPQA
</td>
<td>0
</td>
<td>em
</td>
<td>34.6
</td>
<td>30.4
</td>
<td>39.5
</td>
<td>46.7
</td>
<td>50.7
</td>
</tr>
<tr>
<td rowspan="4" >Code
</td>
<td>HumanEval
</td>
<td>0
</td>
<td>pass@1
</td>
<td>60.4
</td>
<td>72.6
</td>
<td>81.7
</td>
<td>80.5
</td>
<td>89.0
</td>
</tr>
<tr>
<td>MBPP ++ base version
</td>
<td>0
</td>
<td>pass@1
</td>
<td>70.6
</td>
<td>72.8
</td>
<td>82.5
</td>
<td>86.0
</td>
<td>88.6
</td>
</tr>
<tr>
<td>Multipl-E HumanEval
</td>
<td>0
</td>
<td>pass@1
</td>
<td>-
</td>
<td>50.8
</td>
<td>-
</td>
<td>65.5
</td>
<td>75.2
</td>
</tr>
<tr>
<td>Multipl-E MBPP
</td>
<td>0
</td>
<td>pass@1
</td>
<td>-
</td>
<td>52.4
</td>
<td>-
</td>
<td>62.0
</td>
<td>65.7
</td>
</tr>
<tr>
<td rowspan="2" >Math
</td>
<td>GSM-8K (CoT)
</td>
<td>8
</td>
<td>em_maj1@1
</td>
<td>80.6
</td>
<td>84.5
</td>
<td>93.0
</td>
<td>95.1
</td>
<td>96.8
</td>
</tr>
<tr>
<td>MATH (CoT)
</td>
<td>0
</td>
<td>final_em
</td>
<td>29.1
</td>
<td>51.9
</td>
<td>51.0
</td>
<td>68.0
</td>
<td>73.8
</td>
</tr>
<tr>
<td rowspan="4" >Tool Use
</td>
<td>API-Bank
</td>
<td>0
</td>
<td>acc
</td>
<td>48.3
</td>
<td>82.6
</td>
<td>85.1
</td>
<td>90.0
</td>
<td>92.0
</td>
</tr>
<tr>
<td>BFCL
</td>
<td>0
</td>
<td>acc
</td>
<td>60.3
</td>
<td>76.1
</td>
<td>83.0
</td>
<td>84.8
</td>
<td>88.5
</td>
</tr>
<tr>
<td>Gorilla Benchmark API Bench
</td>
<td>0
</td>
<td>acc
</td>
<td>1.7
</td>
<td>8.2
</td>
<td>14.7
</td>
<td>29.7
</td>
<td>35.3
</td>
</tr>
<tr>
<td>Nexus (0-shot)
</td>
<td>0
</td>
<td>macro_avg/acc
</td>
<td>18.1
</td>
<td>38.5
</td>
<td>47.8
</td>
<td>56.7
</td>
<td>58.7
</td>
</tr>
<tr>
<td>Multilingual
</td>
<td>Multilingual MGSM (CoT)
</td>
<td>0
</td>
<td>em
</td>
<td>-
</td>
<td>68.9
</td>
<td>-
</td>
<td>86.9
</td>
<td>91.6
</td>
</tr>
</table>
#### Multilingual benchmarks
<table>
<tr>
<td><strong>Category</strong>
</td>
<td><strong>Benchmark</strong>
</td>
<td><strong>Language</strong>
</td>
<td><strong>Llama 3.1 8B</strong>
</td>
<td><strong>Llama 3.1 70B</strong>
</td>
<td><strong>Llama 3.1 405B</strong>
</td>
</tr>
<tr>
<td rowspan="9" ><strong>General</strong>
</td>
<td rowspan="9" ><strong>MMLU (5-shot, macro_avg/acc)</strong>
</td>
<td>Portuguese
</td>
<td>62.12
</td>
<td>80.13
</td>
<td>84.95
</td>
</tr>
<tr>
<td>Spanish
</td>
<td>62.45
</td>
<td>80.05
</td>
<td>85.08
</td>
</tr>
<tr>
<td>Italian
</td>
<td>61.63
</td>
<td>80.4
</td>
<td>85.04
</td>
</tr>
<tr>
<td>German
</td>
<td>60.59
</td>
<td>79.27
</td>
<td>84.36
</td>
</tr>
<tr>
<td>French
</td>
<td>62.34
</td>
<td>79.82
</td>
<td>84.66
</td>
</tr>
<tr>
<td>Hindi
</td>
<td>50.88
</td>
<td>74.52
</td>
<td>80.31
</td>
</tr>
<tr>
<td>Thai
</td>
<td>50.32
</td>
<td>72.95
</td>
<td>78.21
</td>
</tr>
</table>
## Responsibility & Safety
As part of our Responsible release approach, we followed a three-pronged strategy to managing trust & safety risks:
* Enable developers to deploy helpful, safe and flexible experiences for their target audience and for the use cases supported by Llama.
* Protect developers against adversarial users aiming to exploit Llama capabilities to potentially cause harm.
* Provide protections for the community to help prevent the misuse of our models.
### Responsible deployment
Llama is a foundational technology designed to be used in a variety of use cases, examples on how Meta’s Llama models have been responsibly deployed can be found in our [Community Stories webpage](https://llama.meta.com/community-stories/). Our approach is to build the most helpful models enabling the world to benefit from the technology power, by aligning our model safety for the generic use cases addressing a standard set of harms. Developers are then in the driver seat to tailor safety for their use case, defining their own policy and deploying the models with the necessary safeguards in their Llama systems. Llama 3.1 was developed following the best practices outlined in our Responsible Use Guide, you can refer to the [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/) to learn more.
#### Llama 3.1 instruct
Our main objectives for conducting safety fine-tuning are to provide the research community with a valuable resource for studying the robustness of safety fine-tuning, as well as to offer developers a readily available, safe, and powerful model for various applications to reduce the developer workload to deploy safe AI systems. For more details on the safety mitigations implemented please read the Llama 3 paper.
**Fine-tuning data**
We employ a multi-faceted approach to data collection, combining human-generated data from our vendors with synthetic data to mitigate potential safety risks. We’ve developed many large language model (LLM)-based classifiers that enable us to thoughtfully select high-quality prompts and responses, enhancing data quality control.
**Refusals and Tone**
Building on the work we started with Llama 3, we put a great emphasis on model refusals to benign prompts as well as refusal tone. We included both borderline and adversarial prompts in our safety data strategy, and modified our safety data responses to follow tone guidelines.
#### Llama 3.1 systems
**Large language models, including Llama 3.1, are not designed to be deployed in isolation but instead should be deployed as part of an overall AI system with additional safety guardrails as required.** Developers are expected to deploy system safeguards when building agentic systems. Safeguards are key to achieve the right helpfulness-safety alignment as well as mitigating safety and security risks inherent to the system and any integration of the model or system with external tools.
As part of our responsible release approach, we provide the community with [safeguards](https://llama.meta.com/trust-and-safety/) that developers should deploy with Llama models or other LLMs, including Llama Guard 3, Prompt Guard and Code Shield. All our [reference implementations](https://github.com/meta-llama/llama-agentic-system) demos contain these safeguards by default so developers can benefit from system-level safety out-of-the-box.
#### New capabilities
Note that this release introduces new capabilities, including a longer context window, multilingual inputs and outputs and possible integrations by developers with third party tools. Building with these new capabilities requires specific considerations in addition to the best practices that generally apply across all Generative AI use cases.
**Tool-use**: Just like in standard software development, developers are responsible for the integration of the LLM with the tools and services of their choice. They should define a clear policy for their use case and assess the integrity of the third party services they use to be aware of the safety and security limitations when using this capability. Refer to the Responsible Use Guide for best practices on the safe deployment of the third party safeguards.
**Multilinguality**: Llama 3.1 supports 7 languages in addition to English: French, German, Hindi, Italian, Portuguese, Spanish, and Thai. Llama may be able to output text in other languages than those that meet performance thresholds for safety and helpfulness. We strongly discourage developers from using this model to converse in non-supported languages without implementing finetuning and system controls in alignment with their policies and the best practices shared in the Responsible Use Guide.
### Evaluations
We evaluated Llama models for common use cases as well as specific capabilities. Common use cases evaluations measure safety risks of systems for most commonly built applications including chat bot, coding assistant, tool calls. We built dedicated, adversarial evaluation datasets and evaluated systems composed of Llama models and Llama Guard 3 to filter input prompt and output response. It is important to evaluate applications in context, and we recommend building dedicated evaluation dataset for your use case. Prompt Guard and Code Shield are also available if relevant to the application.
Capability evaluations measure vulnerabilities of Llama models inherent to specific capabilities, for which were crafted dedicated benchmarks including long context, multilingual, tools calls, coding or memorization.
**Red teaming**
For both scenarios, we conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we used the learnings to improve our benchmarks and safety tuning datasets.
We partnered early with subject-matter experts in critical risk areas to understand the nature of these real-world harms and how such models may lead to unintended harm for society. Based on these conversations, we derived a set of adversarial goals for the red team to attempt to achieve, such as extracting harmful information or reprogramming the model to act in a potentially harmful capacity. The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity in addition to multilingual content specialists with background in integrity issues in specific geographic markets.
### Critical and other risks
We specifically focused our efforts on mitigating the following critical risk areas:
**1- CBRNE (Chemical, Biological, Radiological, Nuclear, and Explosive materials) helpfulness**
To assess risks related to proliferation of chemical and biological weapons, we performed uplift testing designed to assess whether use of Llama 3.1 models could meaningfully increase the capabilities of malicious actors to plan or carry out attacks using these types of weapons.
**2. Child Safety**
Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors including the additional languages Llama 3 is trained on. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
**3. Cyber attack enablement**
Our cyber attack uplift study investigated whether LLMs can enhance human capabilities in hacking tasks, both in terms of skill level and speed.
Our attack automation study focused on evaluating the capabilities of LLMs when used as autonomous agents in cyber offensive operations, specifically in the context of ransomware attacks. This evaluation was distinct from previous studies that considered LLMs as interactive assistants. The primary objective was to assess whether these models could effectively function as independent agents in executing complex cyber-attacks without human intervention.
Our study of Llama-3.1-405B’s social engineering uplift for cyber attackers was conducted to assess the effectiveness of AI models in aiding cyber threat actors in spear phishing campaigns. Please read our Llama 3.1 Cyber security whitepaper to learn more.
### Community
Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership on AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
We also set up the [Llama Impact Grants](https://llama.meta.com/llama-impact-grants/) program to identify and support the most compelling applications of Meta’s Llama model for societal benefit across three categories: education, climate and open innovation. The 20 finalists from the hundreds of applications can be found [here](https://llama.meta.com/llama-impact-grants/#finalists).
Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
The core values of Llama 3.1 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3.1 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
But Llama 3.1 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3.1’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3.1 models, developers should perform safety testing and tuning tailored to their specific applications of the model. Please refer to available resources including our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide), [Trust and Safety](https://llama.meta.com/trust-and-safety/) solutions, and other [resources](https://llama.meta.com/docs/get-started/) to learn more about responsible development. |
Mungert/Fin-R1-GGUF | Mungert | 2025-06-15T19:41:46Z | 1,552 | 7 | transformers | [
"transformers",
"gguf",
"text-generation",
"arxiv:2503.16252",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-21T23:17:12Z | ---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">Fin-R1 GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Fin-R1-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Fin-R1-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Fin-R1-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Fin-R1-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Fin-R1-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Fin-R1-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Fin-R1-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Fin-R1-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Fin-R1-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Fin-R1-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Fin-R1-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
<div align="center">
<h1>Fin-R1：通过强化学习驱动的金融推理大模型</h1>
<!-- 徽章部分 -->
[](https://www.apache.org/licenses/LICENSE-2.0)[](https://huggingface.co/SUFE-AIFLM-Lab/Fin-R1)[](https://arxiv.org/abs/2503.16252)
<!-- 语言切换链接 -->
📄 [中文](https://huggingface.co/SUFE-AIFLM-Lab/Fin-R1/blob/main/README.md) | [EN](https://huggingface.co/SUFE-AIFLM-Lab/Fin-R1/blob/main/README_en.md)
</div>
Fin-R1 是一款针对金融领域复杂推理的大型语言模型，由上海财经大学统计与数据科学学院张立文教授与其领衔的金融大语言模型课题组（SUFE-AIFLM-Lab）联合财跃星辰研发并开源发布。该模型以 Qwen2.5-7B-Instruct 为基座，通过高质量的可验证金融问题微调训练，最终表现在多个金融领域基准测试上的表现达到参评模型的SOTA水平。
Code: https://github.com/SUFE-AIFLM-Lab/Fin-R1
## 📌 目录<a name="toc"></a>
- [场景应用](#summary)
- [金融代码](#金融代码)
- [金融计算](#金融计算)
- [英语金融计算](#英语金融计算)
- [金融安全合规](#金融安全合规)
- [智能风控](#智能风控)
- [ESG分析](#ESG分析)
- [总体工作流程](#总体工作流程)
- [数据构建](#data)
- [微调训练](#trainning)
- [模型评测结果](#results)
- [模型使用方法](#use)
- [未来展望](#todo)
- [联系我们](#connection)
## 💡 场景应用 <a name="summary"></a>
Fin-R1 是一款专为金融推理领域设计的大语言模型，采用轻量化的 7B 参数量级架构。在显著降低部署成本的同时，该模型通过在针对金融推理场景的高质量思维链数据上采用 SFT（监督微调）和 RL（强化学习）两阶段训练，为模型在金融领域的应用提供了坚实的理论支撑、业务规则、决策逻辑以及技术实现能力，从而有效提升模型的金融复杂推理能力，为银行、证券、保险以及信托等金融核心业务场景提供有力支持。
## 金融代码
金融代码是指在金融领域中用于实现各种金融模型、算法和分析任务的计算机编程代码，涵盖了从简单的财务计算到复杂的金融衍生品定价、风险评估和投资组合优化等多个方面，以方便金融专业人士进行数据处理、统计分析、数值计算和可视化等工作。
## 金融计算
金融计算是对金融领域的各种问题进行定量分析和计算的过程，其核心在于通过建立数学模型和运用数值方法来解决实际金融问题，可为金融决策提供科学依据，帮助金融机构和投资者更好地管理风险、优化资源配置和提高投资回报率。
## 英语金融计算
英语金融计算强调在跨语言环境下使用英语进行金融模型的构建和计算，并能够以英语撰写金融分析报告和与国际同行进行沟通交流。
## 金融安全合规
金融安全合规聚焦于防范金融犯罪与遵守监管要求，帮助企业建立健全的合规管理体系，定期进行合规检查和审计，确保业务操作符合相关法规要求。
## 智能风控
智能风控利用AI与大数据技术识别和管理金融风险，与传统风控手段相比，智能风控具有更高的效率、准确性和实时性，它通过对海量金融数据的深度挖掘和分析，能够发现潜在的风险模式和异常交易行为，从而及时预警和采取相应的风险控制措施。
## ESG分析
ESG分析通过评估企业在环境（Environmental）、社会（Social）、治理（Governance）的表现，衡量其可持续发展能力，确保投资活动不仅能够获得财务回报，还能促进可持续发展和社会责任的履行。金融机构和企业也通过提升自身的 ESG 绩效，来满足投资者和社会对企业更高的期望和要求。
## 总体工作流程
我们基于 DeepSeek-R1 构建了数据蒸馏框架，并严格按照官方参数设定进行数据处理，采用两阶段数据筛选方法提升金融领域数据质量，生成了SFT数据集和RL数据集。在训练过程中，我们利用Qwen2.5-7B-Instruct，通过监督微调（SFT）和强化学习（RL）训练金融推理大模型 Fin-R1，以提升金融推理任务的准确性和泛化能力。
## 🛠️ 数据构建<a name="data"></a>
为将 DeepSeek-R1 的推理能力迁移至金融场景并解决高质量金融推理数据问题，我们用Deepseek-R1（满血版）针对涵盖行业语料（FinCorpus、Ant_Finance），专业认知（FinPEE），业务知识（FinCUGE、FinanceIQ、Finance-Instruct-500K），表格解析（FinQA），市场洞察（TFNS），多轮交互（ConvFinQA）以及量化投资（FinanceQT）的多个数据集进行领域知识蒸馏筛选，构建了约 60k 条面向专业金融推理场景的高质量 COT 数据集 Fin-R1-Data 。该数据集涵盖中英文金融垂直领域的多维度专业知识，并根据具体任务内容将其分为金融代码、金融专业知识、金融非推理类业务知识和金融推理类业务知识四大模块，可有效支撑银行、基金和证券等多个金融核心场景。本研究构建了基于 Deepseek-R1 的数据蒸馏框架，并创新性提出对思维链进行“答案+推理”双轮质量打分筛选方法，首轮基于规则匹配和 Qwen2.5-72B-Instruct 对答案准确性评分，次轮对推理链的逻辑一致性、术语合规性等推理逻辑进行深度校验以保证数据质量。
### 数据蒸馏
在蒸馏过程中，我们严格依照 [DeepSeek - R1](https://github.com/deepseek-ai/DeepSeek-R1) 官方提供的细节，进行相应设置的数据蒸馏操作。
### 数据筛选
针对金融数据结构的复杂特性采取对思维链进行“答案+推理逻辑”双轮质量打分的创新方式筛选，首轮基于规则匹配和 Qwen2.5-72B-Instruct 对答案准确性评分，次轮对推理链的逻辑一致性、术语合规性等推理逻辑进行深度校验以保证数据质量，每次打分筛选出的数据标注为 good 或 bad 进行区分：
1）答案打分：对于蒸馏得到的数据，针对客观题（如选择题、判断题），采用基于规则的匹配方式，校对蒸馏数据的正确性；对于无法通过规则匹配的结果，利用 Qwen2.5-72B-Instruct 对模型生成的答案以及正确答案进行打分，正确得 1 分，错误得 0 分。
2）推理过程打分：对于经过上一步筛选得到的正确思维链数据，再次利用 Qwen2.5-72B-Instruct 对推理轨迹进行打分，高质量数据得 1 分，低质量数据得 0 分。我们采取了如下几个指标来进行打分：
>
> 1.内部一致性：检查推理过程中的步骤是否一致，并且是否能够逐步逻辑地推导出标准答案。
>
> 2.术语重叠度：检查推理过程中使用的术语与标准答案中的术语的重叠程度。重叠度越高越好。
>
> 3.推理步骤数量：评估推理过程是否包含足够的步骤（至少3步）。
>
> 4.逻辑一致性：确保推理过程中的步骤与标准答案在逻辑上高度一致，并检查是否存在明显的错误或遗漏。
>
> 5.内容多样性：检查推理过程中是否存在大量重复的步骤。
>
> 6.与任务领域的相关性：检查推理过程是否涉及与任务领域相关的内容（任务领域：{task_domain}）。如果推理反映了与任务领域的相关性，则给予更高的评分。
>
> 7.与任务指令的一致性：检查推理过程是否与任务指令高度相关。相关性越高越好。如果推理内容完全符合任务指令，则给予更高的评分。
我们将经过两轮筛选后均标注为good的数据作为高质量的 COT 数据用于 SFT ；而未经过筛选标注为bad的数据则作为推理QA数据用于强化学习（RL）。
### Fin-R1-Data数据分布如下：
Fin-R1-Data 涵盖中英文金融垂直领域的多维度专业知识，并根据具体任务内容将其分为金融代码、金融专业知识、金融非推理类业务知识和金融推理类业务知识四大模块，可有效支撑银行、证券以及信托等多个金融核心场景。
|数据集|数据量|
|-------------|--------|
|ConvFinQA-R1-Distill |7629|
|Finance-Instruct-500K-R1-Distill | 11300 |
|FinCUGE-R1-Distill | 2000 |
|FinQA-R1-Distill | 2948 |
|TFNS-R1-Distill | 2451|
|FinanceIQ-R1-Distill | 2596 |
|FinanceQT-R1-Distill | 152 |
|Ant_Finance-R1-Distill | 1548 |
|FinCorpus-R1-Distill | 29288|
|FinPEE-R1-Distill | 179 |
|总计| 60091 |
## 🚀 微调训练<a name="trainning"></a>
### 两阶段流程
针对金融领域复杂推理任务，我们利用 Qwen2.5-7B-Instruct 进行两阶段微调训练得到金融推理大语言模型 Fin-R1 。首先通过高质量金融推理数据的 SFT (Supervised Fine-Tuning) 帮助模型初步提升金融推理能力，然后在 GRPO（Group Relative Policy Optimization) 算法的基础上结合格式奖励和准确度奖励进行强化学习，以此进一步提升金融推理任务的准确性和泛化能力。
#### 第一阶段----推理能力注入：
针对金融推理任务中的复杂推理，我们第一阶段使用 ConvFinQA 和 FinQA 金融数据集对 Qwen2.5-7B-Instruct 进行了监督微调。经过一轮微调训练，确保模型能够深入理解并处理复杂的金融推理问题。
#### 第二阶段----强化学习优化：
在模型掌握复杂推理技能后，我们采用 GRPO（Group Relative Policy Optimization）算法作为核心框架，以双重奖励机制优化模型输出的格式和准确度，并在此基础上引入了基于模型的验证器（Model-Based Verifier），采用 Qwen2.5-Max 进行答案评估来改进基于正则表达式的奖励可能存在的偏差，生成更加精确可靠的奖励信号，从而提升强化学习的效果和稳定性。
## 🚨 模型评测结果 <a name="results"></a>
我们在覆盖多项金融业务场景的基准测试上对模型进行评估，在评测结果中，只经过指令微调 (SFT) 的模型 Fin-R1-SFT 在金融场景中相较于基础模型已经取得了一定性能提升，但是相比于 DeepSeek-R1 仍有提升空间，我们于是在 Fin-R1-SFT 基础上再进行强化学习训练，结果发现经过指令微调 (SFT) 加强化学习 (RL) 训练的 Fin-R1 以仅 7B 的轻量化参数规模展现出显著的性能优势，达到 75.2 的平均得分位居第二，全面超越参评的同规模模型，同时与行业标杆 DeepSeek-R1 平均分差距仅3.0， 且超越DeepSeek-R1-Distill-Llama-70B（69.2）6.0分。此外 Fin-R1 在聚焦真实金融表格数值推理任务的 FinQA 以及多轮推理交互场景的 ConvFinQA 两大关键任务测试上分别以 76.0 和 85.0 的得分在参评模型中登顶第一，展现出了模型在金融推理场景及金融非推理场景中的强大处理能力。
| Model | Parameters | FinQA | ConvFinQA | Ant_Finance | TFNS | Finance-Instruct-500k | Average |
|------------------------------|------------|--------|-----------|-------------|--------|-------------------------|---------|
| DeepSeek-R1 | 671B | 71.0 | 82.0 | __90.0__ | 78.0 | __70.0__ | __78.2__|
| __Fin-R1__ | 7B |__76.0__| __85.0__ | 81.0 | 71.0 | 62.9 | 75.2 |
| Qwen-2.5-32B-Instruct | 32B | 72.0 | 78.0 | 84.0 | 77.0 | 58.0 | 73.8 |
| DeepSeek-R1-Distill-Qwen-32B | 32B | 70.0 | 72.0 | 87.0 |__79.0__| 54.0 | 72.4 |
| __Fin-R1-SFT__ | 7B | 73.0 | 81.0 | 76.0 | 68.0 | 61.0 | 71.9 |
| Qwen-2.5-14B-Instruct | 14B | 68.0 | 77.0 | 84.0 | 72.0 | 56.0 | 71.4 |
| DeepSeek-R1-Distill-Llama-70B| 70B | 68.0 | 74.0 | 84.0 | 62.0 | 56.0 | 69.2 |
| DeepSeek-R1-Distill-Qwen-14B | 14B | 62.0 | 73.0 | 82.0 | 65.0 | 49.0 | 66.2 |
| Qwen-2.5-7B-Instruct | 7B | 60.0 | 66.0 | 85.0 | 68.0 | 49.0 | 65.6 |
| DeepSeek-R1-Distill-Qwen-7B | 7B | 55.0 | 62.0 | 71.0 | 60.0 | 42.0 | 58.0 |
## 声明及未来展望 <a name="todo"></a>
本项目由上海财经大学统计与数据科学学院金融大语言模型课题组（SUFE-AIFLM-Lab）联合财跃星辰完成。Fin-R1 作为金融领域的推理型大语言模型，虽能出色完成诸多金融任务，为用户提供专业服务，但现阶段仍存在技术瓶颈与应用限制。它提供的建议和分析结果仅供参考，不可等同于专业金融分析师或专家的精准判断。我们诚挚希望用户以批判性思维审视模型输出，结合自身专业知识与经验进行决策。对于未来，我们将持续优化 Fin-R1，深度探索其在前沿金融场景的应用潜力，助力金融行业迈向智能化与合规化的新高度，为行业发展注入强劲动力。
## 📫 联系我们 <a name="connection"></a>
诚邀业界同仁共同探索 AI 与金融深度融合的创新范式，共建智慧金融新生态，并通过邮件与[email protected]联系 |
Mungert/Llama-3.1-Nemotron-Nano-8B-v1-GGUF | Mungert | 2025-06-15T19:41:41Z | 2,163 | 7 | transformers | [
"transformers",
"gguf",
"nvidia",
"llama-3",
"pytorch",
"text-generation",
"en",
"arxiv:2505.00949",
"arxiv:2502.00203",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-21T19:44:49Z | ---
library_name: transformers
license: other
license_name: nvidia-open-model-license
license_link: >-
https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
pipeline_tag: text-generation
language:
- en
tags:
- nvidia
- llama-3
- pytorch
---
# <span style="color: #7FFF7F;">Llama-3.1-Nemotron-Nano-8B-v1 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.1-Nemotron-Nano-8B-v1-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.1-Nemotron-Nano-8B-v1-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.1-Nemotron-Nano-8B-v1-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.1-Nemotron-Nano-8B-v1-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.1-Nemotron-Nano-8B-v1-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.1-Nemotron-Nano-8B-v1-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.1-Nemotron-Nano-8B-v1-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Llama-3.1-Nemotron-Nano-8B-v1
## Model Overview
Llama-3.1-Nemotron-Nano-8B-v1 is a large language model (LLM) which is a derivative of [Meta Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) (AKA the reference model). It is a reasoning model that is post trained for reasoning, human chat preferences, and tasks, such as RAG and tool calling.
Llama-3.1-Nemotron-Nano-8B-v1 is a model which offers a great tradeoff between model accuracy and efficiency. It is created from Llama 3.1 8B Instruct and offers improvements in model accuracy. The model fits on a single RTX GPU and can be used locally. The model supports a context length of 128K.
This model underwent a multi-phase post-training process to enhance both its reasoning and non-reasoning capabilities. This includes a supervised fine-tuning stage for Math, Code, Reasoning, and Tool Calling as well as multiple reinforcement learning (RL) stages using REINFORCE (RLOO) and Online Reward-aware Preference Optimization (RPO) algorithms for both chat and instruction-following. The final model checkpoint is obtained after merging the final SFT and Online RPO checkpoints. Improved using Qwen.
This model is part of the Llama Nemotron Collection. You can find the other model(s) in this family here:
[Llama-3.3-Nemotron-Super-49B-v1](https://huggingface.co/nvidia/Llama-3.3-Nemotron-Super-49B-v1)
This model is ready for commercial use.
## License/Terms of Use
GOVERNING TERMS: Your use of this model is governed by the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). Additional Information: [Llama 3.1 Community License Agreement](https://www.llama.com/llama3_1/license/). Built with Llama.
**Model Developer:** NVIDIA
**Model Dates:** Trained between August 2024 and March 2025
**Data Freshness:** The pretraining data has a cutoff of 2023 per Meta Llama 3.1 8B
## Use Case:
Developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks. Balance of model accuracy and compute efficiency (the model fits on a single RTX GPU and can be used locally).
## Release Date: <br>
3/18/2025 <br>
## References
- [\[2505.00949\] Llama-Nemotron: Efficient Reasoning Models](https://arxiv.org/abs/2505.00949)
- [\[2502.00203\] Reward-aware Preference Optimization: A Unified Mathematical Framework for Model Alignment](https://arxiv.org/abs/2502.00203)
## Model Architecture
**Architecture Type:** Dense decoder-only Transformer model
**Network Architecture:** Llama 3.1 8B Instruct
## Intended use
Llama-3.1-Nemotron-Nano-8B-v1 is a general purpose reasoning and chat model intended to be used in English and coding languages. Other non-English languages (German, French, Italian, Portuguese, Hindi, Spanish, and Thai) are also supported.
# Input:
- **Input Type:** Text
- **Input Format:** String
- **Input Parameters:** One-Dimensional (1D)
- **Other Properties Related to Input:** Context length up to 131,072 tokens
## Output:
- **Output Type:** Text
- **Output Format:** String
- **Output Parameters:** One-Dimensional (1D)
- **Other Properties Related to Output:** Context length up to 131,072 tokens
## Model Version:
1.0 (3/18/2025)
## Software Integration
- **Runtime Engine:** NeMo 24.12 <br>
- **Recommended Hardware Microarchitecture Compatibility:**
- NVIDIA Hopper
- NVIDIA Ampere
## Quick Start and Usage Recommendations:
1. Reasoning mode (ON/OFF) is controlled via the system prompt, which must be set as shown in the example below. All instructions should be contained within the user prompt
2. We recommend setting temperature to `0.6`, and Top P to `0.95` for Reasoning ON mode
3. We recommend using greedy decoding for Reasoning OFF mode
4. We have provided a list of prompts to use for evaluation for each benchmark where a specific template is required
5. The model will include `<think></think>` if no reasoning was necessary in Reasoning ON model, this is expected behaviour
You can try this model out through the preview API, using this link: [Llama-3.1-Nemotron-Nano-8B-v1](https://build.nvidia.com/nvidia/llama-3_1-nemotron-nano-8b-v1).
See the snippet below for usage with Hugging Face Transformers library. Reasoning mode (ON/OFF) is controlled via system prompt. Please see the example below.
Our code requires the transformers package version to be `4.44.2` or higher.
### Example of “Reasoning On:”
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
temperature=0.6,
top_p=0.95,
**model_kwargs
)
# Thinking can be "on" or "off"
thinking = "on"
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}]))
```
### Example of “Reasoning Off:”
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
do_sample=False,
**model_kwargs
)
# Thinking can be "on" or "off"
thinking = "off"
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}]))
```
For some prompts, even though thinking is disabled, the model emergently prefers to think before responding. But if desired, the users can prevent it by pre-filling the assistant response.
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
# Thinking can be "on" or "off"
thinking = "off"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
do_sample=False,
**model_kwargs
)
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}, {"role":"assistant", "content":"<think>\n</think>"}]))
```
## Inference:
**Engine:** Transformers
**Test Hardware:**
- BF16:
- 1x RTX 50 Series GPUs
- 1x RTX 40 Series GPUs
- 1x RTX 30 Series GPUs
- 1x H100-80GB GPU
- 1x A100-80GB GPU
**Preferred/Supported] Operating System(s):** Linux <br>
## Training Datasets
A large variety of training data was used for the post-training pipeline, including manually annotated data and synthetic data.
The data for the multi-stage post-training phases for improvements in Code, Math, and Reasoning is a compilation of SFT and RL data that supports improvements of math, code, general reasoning, and instruction following capabilities of the original Llama instruct model.
Prompts have been sourced from either public and open corpus or synthetically generated. Responses were synthetically generated by a variety of models, with some prompts containing responses for both Reasoning On and Off modes, to train the model to distinguish between two modes.
**Data Collection for Training Datasets:** <br>
* Hybrid: Automated, Human, Synthetic <br>
**Data Labeling for Training Datasets:** <br>
* N/A <br>
## Evaluation Datasets
We used the datasets listed below to evaluate Llama-3.1-Nemotron-Nano-8B-v1.
**Data Collection for Evaluation Datasets:** Hybrid: Human/Synthetic
**Data Labeling for Evaluation Datasets:** Hybrid: Human/Synthetic/Automatic
## Evaluation Results
These results contain both “Reasoning On”, and “Reasoning Off”. We recommend using temperature=`0.6`, top_p=`0.95` for “Reasoning On” mode, and greedy decoding for “Reasoning Off” mode. All evaluations are done with 32k sequence length. We run the benchmarks up to 16 times and average the scores to be more accurate.
> NOTE: Where applicable, a Prompt Template will be provided. While completing benchmarks, please ensure that you are parsing for the correct output format as per the provided prompt in order to reproduce the benchmarks seen below.
### MT-Bench
| Reasoning Mode | Score |
|--------------|------------|
| Reasoning Off | 7.9 |
| Reasoning On | 8.1 |
### MATH500
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 36.6% |
| Reasoning On | 95.4% |
User Prompt Template:
```
"Below is a math question. I want you to reason through the steps and then give a final answer. Your final answer should be in \boxed{}.\nQuestion: {question}"
```
### AIME25
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 0% |
| Reasoning On | 47.1% |
User Prompt Template:
```
"Below is a math question. I want you to reason through the steps and then give a final answer. Your final answer should be in \boxed{}.\nQuestion: {question}"
```
### GPQA-D
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 39.4% |
| Reasoning On | 54.1% |
User Prompt Template:
```
"What is the correct answer to this question: {question}\nChoices:\nA. {option_A}\nB. {option_B}\nC. {option_C}\nD. {option_D}\nLet's think step by step, and put the final answer (should be a single letter A, B, C, or D) into a \boxed{}"
```
### IFEval Average
| Reasoning Mode | Strict:Prompt | Strict:Instruction |
|--------------|------------|------------|
| Reasoning Off | 74.7% | 82.1% |
| Reasoning On | 71.9% | 79.3% |
### BFCL v2 Live
| Reasoning Mode | Score |
|--------------|------------|
| Reasoning Off | 63.9% |
| Reasoning On | 63.6% |
User Prompt Template:
```
<AVAILABLE_TOOLS>{functions}</AVAILABLE_TOOLS>
{user_prompt}
```
### MBPP 0-shot
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 66.1% |
| Reasoning On | 84.6% |
User Prompt Template:
````
You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.
@@ Instruction
Here is the given problem and test examples:
{prompt}
Please use the python programming language to solve this problem.
Please make sure that your code includes the functions from the test samples and that the input and output formats of these functions match the test samples.
Please return all completed codes in one code block.
This code block should be in the following format:
```python
# Your codes here
```
````
## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ [Explainability](explainability.md), [Bias](bias.md), [Safety & Security](safety.md), and [Privacy](privacy.md) Subcards.
Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
## Citation
```
@misc{bercovich2025llamanemotronefficientreasoningmodels,
title={Llama-Nemotron: Efficient Reasoning Models},
author={Akhiad Bercovich and Itay Levy and Izik Golan and Mohammad Dabbah and Ran El-Yaniv and Omri Puny and Ido Galil and Zach Moshe and Tomer Ronen and Najeeb Nabwani and Ido Shahaf and Oren Tropp and Ehud Karpas and Ran Zilberstein and Jiaqi Zeng and Soumye Singhal and Alexander Bukharin and Yian Zhang and Tugrul Konuk and Gerald Shen and Ameya Sunil Mahabaleshwarkar and Bilal Kartal and Yoshi Suhara and Olivier Delalleau and Zijia Chen and Zhilin Wang and David Mosallanezhad and Adi Renduchintala and Haifeng Qian and Dima Rekesh and Fei Jia and Somshubra Majumdar and Vahid Noroozi and Wasi Uddin Ahmad and Sean Narenthiran and Aleksander Ficek and Mehrzad Samadi and Jocelyn Huang and Siddhartha Jain and Igor Gitman and Ivan Moshkov and Wei Du and Shubham Toshniwal and George Armstrong and Branislav Kisacanin and Matvei Novikov and Daria Gitman and Evelina Bakhturina and Jane Polak Scowcroft and John Kamalu and Dan Su and Kezhi Kong and Markus Kliegl and Rabeeh Karimi and Ying Lin and Sanjeev Satheesh and Jupinder Parmar and Pritam Gundecha and Brandon Norick and Joseph Jennings and Shrimai Prabhumoye and Syeda Nahida Akter and Mostofa Patwary and Abhinav Khattar and Deepak Narayanan and Roger Waleffe and Jimmy Zhang and Bor-Yiing Su and Guyue Huang and Terry Kong and Parth Chadha and Sahil Jain and Christine Harvey and Elad Segal and Jining Huang and Sergey Kashirsky and Robert McQueen and Izzy Putterman and George Lam and Arun Venkatesan and Sherry Wu and Vinh Nguyen and Manoj Kilaru and Andrew Wang and Anna Warno and Abhilash Somasamudramath and Sandip Bhaskar and Maka Dong and Nave Assaf and Shahar Mor and Omer Ullman Argov and Scot Junkin and Oleksandr Romanenko and Pedro Larroy and Monika Katariya and Marco Rovinelli and Viji Balas and Nicholas Edelman and Anahita Bhiwandiwalla and Muthu Subramaniam and Smita Ithape and Karthik Ramamoorthy and Yuting Wu and Suguna Varshini Velury and Omri Almog and Joyjit Daw and Denys Fridman and Erick Galinkin and Michael Evans and Katherine Luna and Leon Derczynski and Nikki Pope and Eileen Long and Seth Schneider and Guillermo Siman and Tomasz Grzegorzek and Pablo Ribalta and Monika Katariya and Joey Conway and Trisha Saar and Ann Guan and Krzysztof Pawelec and Shyamala Prayaga and Oleksii Kuchaiev and Boris Ginsburg and Oluwatobi Olabiyi and Kari Briski and Jonathan Cohen and Bryan Catanzaro and Jonah Alben and Yonatan Geifman and Eric Chung and Chris Alexiuk},
year={2025},
eprint={2505.00949},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2505.00949},
}
``` |
Mungert/RWKV7-Goose-World3-2.9B-HF-GGUF | Mungert | 2025-06-15T19:41:38Z | 1,401 | 18 | null | [
"gguf",
"text-generation",
"en",
"zh",
"ja",
"ko",
"fr",
"ar",
"es",
"pt",
"base_model:BlinkDL/rwkv-7-world",
"base_model:quantized:BlinkDL/rwkv-7-world",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-21T03:09:13Z | ---
license: apache-2.0
language:
- en
- zh
- ja
- ko
- fr
- ar
- es
- pt
metrics:
- accuracy
base_model:
- BlinkDL/rwkv-7-world
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">RWKV7-Goose-World3-2.9B-HF GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `RWKV7-Goose-World3-2.9B-HF-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `RWKV7-Goose-World3-2.9B-HF-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `RWKV7-Goose-World3-2.9B-HF-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `RWKV7-Goose-World3-2.9B-HF-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `RWKV7-Goose-World3-2.9B-HF-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `RWKV7-Goose-World3-2.9B-HF-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `RWKV7-Goose-World3-2.9B-HF-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `RWKV7-Goose-World3-2.9B-HF-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `RWKV7-Goose-World3-2.9B-HF-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `RWKV7-Goose-World3-2.9B-HF-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `RWKV7-Goose-World3-2.9B-HF-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# rwkv7-2.9B-world
<!-- Provide a quick summary of what the model is/does. -->
This is RWKV-7 model under flash-linear attention format.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Bo Peng, Yu Zhang, Songlin Yang, Ruichong Zhang
- **Funded by:** RWKV Project (Under LF AI & Data Foundation)
- **Model type:** RWKV7
- **Language(s) (NLP):** English
- **License:** Apache-2.0
- **Parameter count:** 2.9B
- **Tokenizer:** RWKV World tokenizer
- **Vocabulary size:** 65,536
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** https://github.com/fla-org/flash-linear-attention ; https://github.com/BlinkDL/RWKV-LM
- **Paper:** With in Progress
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
Install `flash-linear-attention` and the latest version of `transformers` before using this model:
```bash
pip install git+https://github.com/fla-org/flash-linear-attention
pip install 'transformers>=4.48.0'
```
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
You can use this model just as any other HuggingFace models:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('fla-hub/rwkv7-2.9B-world', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('fla-hub/rwkv7-2.9B-world', trust_remote_code=True)
model = model.cuda()
prompt = "What is a large language model?"
messages = [
{"role": "user", "content": "Who are you?"},
{"role": "assistant", "content": "I am a GPT-3 based model."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=1024,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)[0]
print(response)
```
### Training Data
This model is trained on the World v3 with a total of 3.119 trillion tokens.
#### Training Hyperparameters
- **Training regime:** bfloat16, lr 4e-4 to 1e-5 "delayed" cosine decay, wd 0.1 (with increasing batch sizes during the middle)
- **Final Loss:** 1.8745
- **Token Count:** 3.119 trillion
## FAQ
Q: safetensors metadata is none.
A: upgrade transformers to >=4.48.0: `pip install 'transformers>=4.48.0'` |
Mungert/EXAONE-Deep-7.8B-GGUF | Mungert | 2025-06-15T19:41:09Z | 1,344 | 5 | transformers | [
"transformers",
"gguf",
"lg-ai",
"exaone",
"exaone-deep",
"text-generation",
"en",
"ko",
"arxiv:2503.12524",
"base_model:LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
"base_model:finetune:LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-19T21:27:57Z | ---
base_model: LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct
base_model_relation: finetune
license: other
license_name: exaone
license_link: LICENSE
language:
- en
- ko
tags:
- lg-ai
- exaone
- exaone-deep
pipeline_tag: text-generation
library_name: transformers
---
# <span style="color: #7FFF7F;">EXAONE-Deep-7.8B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `EXAONE-Deep-7.8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `EXAONE-Deep-7.8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `EXAONE-Deep-7.8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `EXAONE-Deep-7.8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `EXAONE-Deep-7.8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `EXAONE-Deep-7.8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `EXAONE-Deep-7.8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `EXAONE-Deep-7.8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `EXAONE-Deep-7.8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `EXAONE-Deep-7.8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `EXAONE-Deep-7.8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# EXAONE-Deep-7.8B
## Introduction
We introduce EXAONE Deep, which exhibits superior capabilities in various reasoning tasks including math and coding benchmarks, ranging from 2.4B to 32B parameters developed and released by LG AI Research. Evaluation results show that 1) EXAONE Deep **2.4B** outperforms other models of comparable size, 2) EXAONE Deep **7.8B** outperforms not only open-weight models of comparable scale but also a proprietary reasoning model OpenAI o1-mini, and 3) EXAONE Deep **32B** demonstrates competitive performance against leading open-weight models.
For more details, please refer to our [documentation](https://arxiv.org/abs/2503.12524), [blog](https://www.lgresearch.ai/news/view?seq=543) and [GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep).
<p align="center">
<img src="assets/exaone_deep_overall_performance.png", width="100%", style="margin: 40 auto;">
This repository contains the reasoning 7.8B language model with the following features:
- Number of Parameters (without embeddings): 6.98B
- Number of Layers: 32
- Number of Attention Heads: GQA with 32 Q-heads and 8 KV-heads
- Vocab Size: 102,400
- Context Length: 32,768 tokens
## Quickstart
We recommend to use `transformers` v4.43.1 or later.
Here is the code snippet to run conversational inference with the model:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread
model_name = "LGAI-EXAONE/EXAONE-Deep-7.8B"
streaming = True # choose the streaming option
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.bfloat16,
trust_remote_code=True,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
# Choose your prompt:
# Math example (AIME 2024)
prompt = r"""Let $x,y$ and $z$ be positive real numbers that satisfy the following system of equations:
\[\log_2\left({x \over yz}\right) = {1 \over 2}\]\[\log_2\left({y \over xz}\right) = {1 \over 3}\]\[\log_2\left({z \over xy}\right) = {1 \over 4}\]
Then the value of $\left|\log_2(x^4y^3z^2)\right|$ is $\tfrac{m}{n}$ where $m$ and $n$ are relatively prime positive integers. Find $m+n$.
Please reason step by step, and put your final answer within \boxed{}."""
# Korean MCQA example (CSAT Math 2025)
prompt = r"""Question : $a_1 = 2$인 수열 $\{a_n\}$과 $b_1 = 2$인 등차수열 $\{b_n\}$이 모든 자연수 $n$에 대하여\[\sum_{k=1}^{n} \frac{a_k}{b_{k+1}} = \frac{1}{2} n^2\]을 만족시킬 때, $\sum_{k=1}^{5} a_k$의 값을 구하여라.
Options :
A) 120
B) 125
C) 130
D) 135
E) 140
Please reason step by step, and you should write the correct option alphabet (A, B, C, D or E) within \\boxed{}."""
messages = [
{"role": "user", "content": prompt}
]
input_ids = tokenizer.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_tensors="pt"
)
if streaming:
streamer = TextIteratorStreamer(tokenizer)
thread = Thread(target=model.generate, kwargs=dict(
input_ids=input_ids.to("cuda"),
eos_token_id=tokenizer.eos_token_id,
max_new_tokens=32768,
do_sample=True,
temperature=0.6,
top_p=0.95,
streamer=streamer
))
thread.start()
for text in streamer:
print(text, end="", flush=True)
else:
output = model.generate(
input_ids.to("cuda"),
eos_token_id=tokenizer.eos_token_id,
max_new_tokens=32768,
do_sample=True,
temperature=0.6,
top_p=0.95,
)
print(tokenizer.decode(output[0]))
```
> ### Note
> The EXAONE Deep models are trained with an optimized configuration,
> so we recommend following the [Usage Guideline](#usage-guideline) section to achieve optimal performance.
## Evaluation
The following table shows the evaluation results of reasoning tasks such as math and coding. The full evaluation results can be found in the [documentation](https://arxiv.org/abs/2503.12524).
<table>
<tr>
<th>Models</th>
<th>MATH-500 (pass@1)</th>
<th>AIME 2024 (pass@1 / cons@64)</th>
<th>AIME 2025 (pass@1 / cons@64)</th>
<th>CSAT Math 2025 (pass@1)</th>
<th>GPQA Diamond (pass@1)</th>
<th>Live Code Bench (pass@1)</th>
</tr>
<tr>
<td>EXAONE Deep 32B</td>
<td>95.7</td>
<td>72.1 / <strong>90.0</strong></td>
<td>65.8 / <strong>80.0</strong></td>
<td><strong>94.5</strong></td>
<td>66.1</td>
<td>59.5</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-32B</td>
<td>94.3</td>
<td>72.6 / 83.3</td>
<td>55.2 / 73.3</td>
<td>84.1</td>
<td>62.1</td>
<td>57.2</td>
</tr>
<tr>
<td>QwQ-32B</td>
<td>95.5</td>
<td>79.5 / 86.7</td>
<td><strong>67.1</strong> / 76.7</td>
<td>94.4</td>
<td>63.3</td>
<td>63.4</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Llama-70B</td>
<td>94.5</td>
<td>70.0 / 86.7</td>
<td>53.9 / 66.7</td>
<td>88.8</td>
<td>65.2</td>
<td>57.5</td>
</tr>
<tr>
<td>DeepSeek-R1 (671B)</td>
<td><strong>97.3</strong></td>
<td><strong>79.8</strong> / 86.7</td>
<td>66.8 / <strong>80.0</strong></td>
<td>89.9</td>
<td><strong>71.5</strong></td>
<td><strong>65.9</strong></td>
</tr>
<tr>
<th colspan="7" height="30px"></th>
</tr>
<tr>
<td>EXAONE Deep 7.8B</td>
<td><strong>94.8</strong></td>
<td><strong>70.0</strong> / <strong>83.3</strong></td>
<td><strong>59.6</strong> / <strong>76.7</strong></td>
<td><strong>89.9</strong></td>
<td><strong>62.6</strong></td>
<td><strong>55.2</strong></td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-7B</td>
<td>92.8</td>
<td>55.5 / <strong>83.3</strong></td>
<td>38.5 / 56.7</td>
<td>79.7</td>
<td>49.1</td>
<td>37.6</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Llama-8B</td>
<td>89.1</td>
<td>50.4 / 80.0</td>
<td>33.6 / 53.3</td>
<td>74.1</td>
<td>49.0</td>
<td>39.6</td>
</tr>
<tr>
<td>OpenAI o1-mini</td>
<td>90.0</td>
<td>63.6 / 80.0</td>
<td>54.8 / 66.7</td>
<td>84.4</td>
<td>60.0</td>
<td>53.8</td>
</tr>
<tr>
<th colspan="7" height="30px"></th>
</tr>
<tr>
<td>EXAONE Deep 2.4B</td>
<td><strong>92.3</strong></td>
<td><strong>52.5</strong> / <strong>76.7</strong></td>
<td><strong>47.9</strong> / <strong>73.3</strong></td>
<td><strong>79.2</strong></td>
<td><strong>54.3</strong></td>
<td><strong>46.6</strong></td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-1.5B</td>
<td>83.9</td>
<td>28.9 / 52.7</td>
<td>23.9 / 36.7</td>
<td>65.6</td>
<td>33.8</td>
<td>16.9</td>
</tr>
</table>
## Deployment
EXAONE Deep models can be inferred in the various frameworks, such as:
- `TensorRT-LLM`
- `vLLM`
- `SGLang`
- `llama.cpp`
- `Ollama`
- `LM-Studio`
Please refer to our [EXAONE Deep GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep) for more details about the inference frameworks.
## Quantization
We provide the pre-quantized EXAONE Deep models with **AWQ** and several quantization types in **GGUF** format. Please refer to our [EXAONE Deep collection](https://huggingface.co/collections/LGAI-EXAONE/exaone-deep-67d119918816ec6efa79a4aa) to find corresponding quantized models.
## Usage Guideline
To achieve the expected performance, we recommend using the following configurations:
1. Ensure the model starts with `<thought>\n` for reasoning steps. The model's output quality may be degraded when you omit it. You can easily apply this feature by using `tokenizer.apply_chat_template()` with `add_generation_prompt=True`. Please check the example code on [Quickstart](#quickstart) section.
2. The reasoning steps of EXAONE Deep models enclosed by `<thought>\n...\n</thought>` usually have lots of tokens, so previous reasoning steps may be necessary to be removed in multi-turn situation. The provided tokenizer handles this automatically.
3. Avoid using system prompt, and build the instruction on the user prompt.
4. Additional instructions help the models reason more deeply, so that the models generate better output.
- For math problems, the instructions **"Please reason step by step, and put your final answer within \boxed{}."** are helpful.
- For more information on our evaluation setting including prompts, please refer to our [Documentation](https://arxiv.org/abs/2503.12524).
5. In our evaluation, we use `temperature=0.6` and `top_p=0.95` for generation.
6. When evaluating the models, it is recommended to test multiple times to assess the expected performance accurately.
## Limitation
The EXAONE language model has certain limitations and may occasionally generate inappropriate responses. The language model generates responses based on the output probability of tokens, and it is determined during learning from training data. While we have made every effort to exclude personal, harmful, and biased information from the training data, some problematic content may still be included, potentially leading to undesirable responses. Please note that the text generated by EXAONE language model does not reflects the views of LG AI Research.
- Inappropriate answers may be generated, which contain personal, harmful or other inappropriate information.
- Biased responses may be generated, which are associated with age, gender, race, and so on.
- The generated responses rely heavily on statistics from the training data, which can result in the generation of
semantically or syntactically incorrect sentences.
- Since the model does not reflect the latest information, the responses may be false or contradictory.
LG AI Research strives to reduce potential risks that may arise from EXAONE language models. Users are not allowed
to engage in any malicious activities (e.g., keying in illegal information) that may induce the creation of inappropriate
outputs violating LG AI’s ethical principles when using EXAONE language models.
## License
The model is licensed under [EXAONE AI Model License Agreement 1.1 - NC](./LICENSE)
## Citation
```
@article{exaone-deep,
title={EXAONE Deep: Reasoning Enhanced Language Models},
author={{LG AI Research}},
journal={arXiv preprint arXiv:2503.12524},
year={2025}
}
```
## Contact
LG AI Research Technical Support: [email protected] |
sinha-mayank-900/distilhubert-finetuned-gtzan | sinha-mayank-900 | 2025-06-15T19:41:02Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"hubert",
"audio-classification",
"generated_from_trainer",
"dataset:marsyas/gtzan",
"base_model:ntu-spml/distilhubert",
"base_model:finetune:ntu-spml/distilhubert",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"region:us"
] | audio-classification | 2025-06-15T15:07:40Z | ---
library_name: transformers
license: apache-2.0
base_model: ntu-spml/distilhubert
tags:
- generated_from_trainer
datasets:
- marsyas/gtzan
metrics:
- accuracy
model-index:
- name: distilhubert-finetuned-gtzan
results:
- task:
name: Audio Classification
type: audio-classification
dataset:
name: GTZAN
type: marsyas/gtzan
config: all
split: train
args: all
metrics:
- name: Accuracy
type: accuracy
value: 0.86
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilhubert-finetuned-gtzan
This model is a fine-tuned version of [ntu-spml/distilhubert](https://huggingface.co/ntu-spml/distilhubert) on the GTZAN dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6759
- Accuracy: 0.86
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine_with_restarts
- lr_scheduler_warmup_ratio: 0.15
- num_epochs: 30
### Training results
| Training Loss | Epoch | Step | Accuracy | Validation Loss |
|:-------------:|:-----:|:----:|:--------:|:---------------:|
| 2.2385 | 1.0 | 50 | 0.34 | 2.2256 |
| 1.7883 | 2.0 | 100 | 0.545 | 1.7487 |
| 1.4788 | 3.0 | 150 | 0.68 | 1.4187 |
| 1.1262 | 4.0 | 200 | 0.69 | 1.1004 |
| 0.8664 | 5.0 | 250 | 0.735 | 0.9532 |
| 0.7772 | 6.0 | 300 | 0.745 | 0.8106 |
| 0.4455 | 7.0 | 350 | 0.81 | 0.7057 |
| 0.3719 | 8.0 | 400 | 0.815 | 0.6467 |
| 0.3716 | 9.0 | 450 | 0.805 | 0.6164 |
| 0.223 | 10.0 | 500 | 0.825 | 0.5887 |
| 0.1382 | 11.0 | 550 | 0.83 | 0.5941 |
| 0.0729 | 12.0 | 600 | 0.84 | 0.5911 |
| 0.0518 | 13.0 | 650 | 0.84 | 0.6116 |
| 0.0388 | 14.0 | 700 | 0.835 | 0.6217 |
| 0.0304 | 15.0 | 750 | 0.84 | 0.6340 |
| 0.0266 | 16.0 | 800 | 0.84 | 0.6407 |
| 0.026 | 17.0 | 850 | 0.85 | 0.6428 |
| 0.0238 | 18.0 | 900 | 0.84 | 0.6457 |
| 0.0244 | 19.0 | 950 | 0.85 | 0.6457 |
| 0.0278 | 20.0 | 1000 | 0.845 | 0.6466 |
| 0.0676 | 19.0 | 1026 | 0.6300 | 0.8533 |
| 0.0271 | 20.0 | 1080 | 0.6714 | 0.8467 |
| 0.0165 | 21.0 | 1134 | 0.6385 | 0.8533 |
| 0.0158 | 22.0 | 1188 | 0.6895 | 0.8667 |
| 0.0292 | 23.0 | 1242 | 0.6982 | 0.86 |
| 0.0232 | 24.0 | 1296 | 0.6870 | 0.86 |
| 0.0099 | 25.0 | 1350 | 0.6774 | 0.8667 |
| 0.0104 | 26.0 | 1404 | 0.6821 | 0.86 |
| 0.0101 | 27.0 | 1458 | 0.6773 | 0.86 |
| 0.01 | 28.0 | 1512 | 0.6790 | 0.86 |
| 0.0097 | 29.0 | 1566 | 0.6779 | 0.86 |
| 0.0093 | 30.0 | 1620 | 0.6759 | 0.86 |
### Framework versions
- Transformers 4.52.4
- Pytorch 2.7.1+cu126
- Datasets 3.6.0
- Tokenizers 0.21.1
|
Mungert/granite-3.2-8b-instruct-GGUF | Mungert | 2025-06-15T19:41:01Z | 619 | 4 | transformers | [
"transformers",
"gguf",
"language",
"granite-3.2",
"text-generation",
"arxiv:0000.00000",
"base_model:ibm-granite/granite-3.1-8b-instruct",
"base_model:quantized:ibm-granite/granite-3.1-8b-instruct",
"license:apache-2.0",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-19T03:28:48Z | ---
pipeline_tag: text-generation
inference: false
license: apache-2.0
library_name: transformers
tags:
- language
- granite-3.2
base_model:
- ibm-granite/granite-3.1-8b-instruct
---
# <span style="color: #7FFF7F;">granite-3.2-8b-instruct GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `granite-3.2-8b-instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `granite-3.2-8b-instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `granite-3.2-8b-instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `granite-3.2-8b-instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `granite-3.2-8b-instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `granite-3.2-8b-instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `granite-3.2-8b-instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `granite-3.2-8b-instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `granite-3.2-8b-instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `granite-3.2-8b-instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `granite-3.2-8b-instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# Granite-3.2-8B-Instruct
**Model Summary:**
Granite-3.2-8B-Instruct is an 8-billion-parameter, long-context AI model fine-tuned for thinking capabilities. Built on top of [Granite-3.1-8B-Instruct](https://huggingface.co/ibm-granite/granite-3.1-8b-instruct), it has been trained using a mix of permissively licensed open-source datasets and internally generated synthetic data designed for reasoning tasks. The model allows controllability of its thinking capability, ensuring it is applied only when required.
- **Developers:** Granite Team, IBM
- **Website**: [Granite Docs](https://www.ibm.com/granite/docs/)
- **Release Date**: February 26th, 2025
- **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
**Supported Languages:**
English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. However, users may finetune this Granite model for languages beyond these 12 languages.
**Intended Use:**
This model is designed to handle general instruction-following tasks and can be integrated into AI assistants across various domains, including business applications.
**Capabilities**
* **Thinking**
* Summarization
* Text classification
* Text extraction
* Question-answering
* Retrieval Augmented Generation (RAG)
* Code related tasks
* Function-calling tasks
* Multilingual dialog use cases
* Long-context tasks including long document/meeting summarization, long document QA, etc.
**Generation:**
This is a simple example of how to use Granite-3.2-8B-Instruct model.
Install the following libraries:
```shell
pip install torch torchvision torchaudio
pip install accelerate
pip install transformers
```
Then, copy the snippet from the section that is relevant for your use case.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
import torch
model_path="ibm-granite/granite-3.2-8b-instruct"
device="cuda"
model = AutoModelForCausalLM.from_pretrained(
model_path,
device_map=device,
torch_dtype=torch.bfloat16,
)
tokenizer = AutoTokenizer.from_pretrained(
model_path
)
conv = [{"role": "user", "content":"You have 10 liters of a 30% acid solution. How many liters of a 70% acid solution must be added to achieve a 50% acid mixture?"}]
input_ids = tokenizer.apply_chat_template(conv, return_tensors="pt", thinking=True, return_dict=True, add_generation_prompt=True).to(device)
set_seed(42)
output = model.generate(
**input_ids,
max_new_tokens=8192,
)
prediction = tokenizer.decode(output[0, input_ids["input_ids"].shape[1]:], skip_special_tokens=True)
print(prediction)
```
**Example Outputs**
- thinking=True
```md
Here is my thought process:
1. We know we start with 10 liters of a 30% acid solution, so there's 3 liters of pure acid in it (10 liters * 0.3).
2. We need to determine how much of the 70% acid solution to add such that the final mixture is 50% acid.
3. Let's say x liters of the 70% acid solution are added. This will contribute 0.7x liters of pure acid.
4. The total volume of the new mixture will be 10 + x liters, and the total amount of pure acid will be 3 + 0.7x liters.
5. We need to set up an equation where (3 + 0.7x) / (10 + x) = 0.5 (since we want a 50% acid mixture).
6. Solving this equation for x will give us the volume of 70% solution needed.
Here is my response:
To solve this, we need to set up an equation based on the concentration of acid before and after mixing.
1. **Initial Acid Content**:
- We start with 10 liters of a 30% acid solution.
- Pure acid in this = 10 * 0.3 = 3 liters.
2. **Variable for Addition**:
- Let x be the liters of 70% acid solution to be added.
- Pure acid in x liters of 70% solution = 0.7x liters.
3. **Final Mixture Concentration**:
- Total volume of mixture = 10 + x liters.
- Total pure acid in the mixture = 3 (from initial solution) + 0.7x (from added solution).
4. **Setting up the equation for 50% acid concentration**:
- We require that the ratio of pure acid to total volume is 0.5 (50%).
- Therefore, the equation is:
\[
\frac{3 + 0.7x}{10 + x} = 0.5
\]
5. **Solving the equation**:
Multiply both sides by (10 + x) to clear the fraction:
\[
3 + 0.7x = 0.5(10 + x)
\]
Expanding the right side:
\[
3 + 0.7x = 5 + 0.5x
\]
Subtract 0.5x from both sides:
\[
3 + 0.2x = 5
\]
Subtract 3 from both sides:
\[
0.2x = 2
\]
Divide by 0.2:
\[
x = 10
\]
So, 10 liters of the 70% acid solution must be added to the initial 10 liters of 30% solution to get a 50% acid mixture.
Let's verify:
- Initial acid: 3 liters
- Acid from added solution: 0.7 * 10 = 7 liters
- Total acid: 3 + 7 = 10 liters
- Total volume: 10 + 10 = 20 liters
- Acid percentage: \(\frac{10}{20} = 0.5 = 50\%\)
This confirms our solution.
Final answer: 10 liters of 70% acid solution should be added.
```
- thinking=False
```md
Let's denote the amount of 70% acid solution to be added as x liters.
We start with 10 liters of a 30% acid solution, which means there are 0.3 * 10 = 3 liters of pure acid in it.
We want to end up with a 50% acid solution after adding x liters of 70% acid solution. The total volume of the mixture will be 10 + x liters, and we want half of this volume to be pure acid.
So, the amount of pure acid in the final mixture should be 0.5 * (10 + x) liters.
The amount of pure acid in the final mixture comes from two sources: the original 30% solution and the added 70% solution. So, we can set up the following equation:
3 (from the original solution) + 0.7x (from the added solution) = 0.5 * (10 + x)
Now, let's solve for x:
3 + 0.7x = 5 + 0.5x
0.7x - 0.5x = 5 - 3
0.2x = 2
x = 2 / 0.2
x = 10
So, you need to add 10 liters of a 70% acid solution to the 10 liters of a 30% acid solution to get a 50% acid mixture.
```
**Evaluation Results:**
<table>
<thead>
<tr>
<th style="text-align:left; background-color: #001d6c; color: white;">Models</th>
<th style="text-align:center; background-color: #001d6c; color: white;">ArenaHard</th>
<th style="text-align:center; background-color: #001d6c; color: white;">Alpaca-Eval-2</th>
<th style="text-align:center; background-color: #001d6c; color: white;">MMLU</th>
<th style="text-align:center; background-color: #001d6c; color: white;">PopQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">TruthfulQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">BigBenchHard</th>
<th style="text-align:center; background-color: #001d6c; color: white;">DROP</th>
<th style="text-align:center; background-color: #001d6c; color: white;">GSM8K</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval+</th>
<th style="text-align:center; background-color: #001d6c; color: white;">IFEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">AttaQ</th>
</tr></thead>
<tbody>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Llama-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">36.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">27.22</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">69.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">52.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">72.66</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.48</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.24</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.32</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">80.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.10</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.43</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">DeepSeek-R1-Distill-Llama-8B</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">17.17</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">21.85</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">45.80</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">13.25</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">47.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.71</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">44.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">72.18</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.54</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">62.91</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.50</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.87</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Qwen-2.5-7B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">25.44</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">30.34</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.30</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">18.12</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">63.06</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">70.40</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">54.71</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">84.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">93.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.91</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.90</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">81.90</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">DeepSeek-R1-Distill-Qwen-7B</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">10.36</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">15.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.72</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">9.94</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">47.14</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.04</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.76</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.47</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.89</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.10</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.45</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">37.58</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">30.34</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.77</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.7</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.84</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">68.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.78</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.63</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">73.20</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.73</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.1-2B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">23.3</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">27.17</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">57.11</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">20.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">54.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">18.68</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.45</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">75.26</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">63.59</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">84.7</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.2-2B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">24.86</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">34.51</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">57.18</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">20.56</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.8</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">52.27</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">21.12</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.02</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">80.13</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">73.39</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.23</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.2-8B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">55.25</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.19</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.04</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.92</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">64.77</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.95</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">81.65</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.72</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.31</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.42</td>
</tr>
</tbody></table>
**Training Data:**
Overall, our training data is largely comprised of two key sources: (1) publicly available datasets with permissive license, (2) internal synthetically generated data targeted to enhance reasoning capabilites.
<!-- A detailed attribution of datasets can be found in [Granite 3.2 Technical Report (coming soon)](#), and [Accompanying Author List](https://github.com/ibm-granite/granite-3.0-language-models/blob/main/author-ack.pdf). -->
**Infrastructure:**
We train Granite-3.2-8B-Instruct using IBM's super computing cluster, Blue Vela, which is outfitted with NVIDIA H100 GPUs. This cluster provides a scalable and efficient infrastructure for training our models over thousands of GPUs.
**Ethical Considerations and Limitations:**
Granite-3.2-8B-Instruct builds upon Granite-3.1-8B-Instruct, leveraging both permissively licensed open-source and select proprietary data for enhanced performance. Since it inherits its foundation from the previous model, all ethical considerations and limitations applicable to [Granite-3.1-8B-Instruct](https://huggingface.co/ibm-granite/granite-3.1-8b-instruct) remain relevant.
**Resources**
- ⭐️ Learn about the latest updates with Granite: https://www.ibm.com/granite
- 📄 Get started with tutorials, best practices, and prompt engineering advice: https://www.ibm.com/granite/docs/
- 💡 Learn about the latest Granite learning resources: https://ibm.biz/granite-learning-resources
<!-- ## Citation
```
@misc{granite-models,
author = {author 1, author2, ...},
title = {},
journal = {},
volume = {},
year = {2024},
url = {https://arxiv.org/abs/0000.00000},
}
``` --> |
Mungert/Mistral-Small-3.1-24B-Instruct-2503-GGUF | Mungert | 2025-06-15T19:40:56Z | 12,272 | 8 | vllm | [
"vllm",
"gguf",
"image-text-to-text",
"en",
"fr",
"de",
"es",
"pt",
"it",
"ja",
"ko",
"ru",
"zh",
"ar",
"fa",
"id",
"ms",
"ne",
"pl",
"ro",
"sr",
"sv",
"tr",
"uk",
"vi",
"hi",
"bn",
"base_model:mistralai/Mistral-Small-3.1-24B-Base-2503",
"base_model:quantized:mistralai/Mistral-Small-3.1-24B-Base-2503",
"license:apache-2.0",
"region:us",
"imatrix",
"conversational"
] | image-text-to-text | 2025-03-19T01:11:29Z | ---
language:
- en
- fr
- de
- es
- pt
- it
- ja
- ko
- ru
- zh
- ar
- fa
- id
- ms
- ne
- pl
- ro
- sr
- sv
- tr
- uk
- vi
- hi
- bn
license: apache-2.0
library_name: vllm
inference: false
base_model:
- mistralai/Mistral-Small-3.1-24B-Base-2503
extra_gated_description: >-
If you want to learn more about how we process your personal data, please read
our <a href="https://mistral.ai/terms/">Privacy Policy</a>.
pipeline_tag: image-text-to-text
---
# <span style="color: #7FFF7F;">Mistral-Small-3.1-24B-Instruct-2503 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`92ecdcc0`](https://github.com/ggerganov/llama.cpp/commit/92ecdcc06a4c405a415bcaa0cb772bc560aa23b1).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Mistral-Small-3.1-24B-Instruct-2503-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Mistral-Small-3.1-24B-Instruct-2503-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Mistral-Small-3.1-24B-Instruct-2503-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Mistral-Small-3.1-24B-Instruct-2503-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Mistral-Small-3.1-24B-Instruct-2503-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Mistral-Small-3.1-24B-Instruct-2503-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Mistral-Small-3.1-24B-Instruct-2503-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Mistral-Small-3.1-24B-Instruct-2503-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Mistral-Small-3.1-24B-Instruct-2503-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Mistral-Small-3.1-24B-Instruct-2503-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Mistral-Small-3.1-24B-Instruct-2503-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Model Card for Mistral-Small-3.1-24B-Instruct-2503
Building upon Mistral Small 3 (2501), Mistral Small 3.1 (2503) **adds state-of-the-art vision understanding** and enhances **long context capabilities up to 128k tokens** without compromising text performance.
With 24 billion parameters, this model achieves top-tier capabilities in both text and vision tasks.
This model is an instruction-finetuned version of: [Mistral-Small-3.1-24B-Base-2503](https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Base-2503).
Mistral Small 3.1 can be deployed locally and is exceptionally "knowledge-dense," fitting within a single RTX 4090 or a 32GB RAM MacBook once quantized.
It is ideal for:
- Fast-response conversational agents.
- Low-latency function calling.
- Subject matter experts via fine-tuning.
- Local inference for hobbyists and organizations handling sensitive data.
- Programming and math reasoning.
- Long document understanding.
- Visual understanding.
For enterprises requiring specialized capabilities (increased context, specific modalities, domain-specific knowledge, etc.), we will release commercial models beyond what Mistral AI contributes to the community.
Learn more about Mistral Small 3.1 in our [blog post](https://mistral.ai/news/mistral-small-3-1/).
## Key Features
- **Vision:** Vision capabilities enable the model to analyze images and provide insights based on visual content in addition to text.
- **Multilingual:** Supports dozens of languages, including English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Malay, Nepali, Polish, Portuguese, Romanian, Russian, Serbian, Spanish, Swedish, Turkish, Ukrainian, Vietnamese, Arabic, Bengali, Chinese, Farsi.
- **Agent-Centric:** Offers best-in-class agentic capabilities with native function calling and JSON outputting.
- **Advanced Reasoning:** State-of-the-art conversational and reasoning capabilities.
- **Apache 2.0 License:** Open license allowing usage and modification for both commercial and non-commercial purposes.
- **Context Window:** A 128k context window.
- **System Prompt:** Maintains strong adherence and support for system prompts.
- **Tokenizer:** Utilizes a Tekken tokenizer with a 131k vocabulary size.
## Benchmark Results
When available, we report numbers previously published by other model providers, otherwise we re-evaluate them using our own evaluation harness.
### Pretrain Evals
| Model | MMLU (5-shot) | MMLU Pro (5-shot CoT) | TriviaQA | GPQA Main (5-shot CoT)| MMMU |
|--------------------------------|---------------|-----------------------|------------|-----------------------|-----------|
| **Small 3.1 24B Base** | **81.01%** | **56.03%** | 80.50% | **37.50%** | **59.27%**|
| Gemma 3 27B PT | 78.60% | 52.20% | **81.30%** | 24.30% | 56.10% |
### Instruction Evals
#### Text
| Model | MMLU | MMLU Pro (5-shot CoT) | MATH | GPQA Main (5-shot CoT) | GPQA Diamond (5-shot CoT )| MBPP | HumanEval | SimpleQA (TotalAcc)|
|--------------------------------|-----------|-----------------------|------------------------|------------------------|---------------------------|-----------|-----------|--------------------|
| **Small 3.1 24B Instruct** | 80.62% | 66.76% | 69.30% | **44.42%** | **45.96%** | 74.71% | **88.41%**| **10.43%** |
| Gemma 3 27B IT | 76.90% | **67.50%** | **89.00%** | 36.83% | 42.40% | 74.40% | 87.80% | 10.00% |
| GPT4o Mini | **82.00%**| 61.70% | 70.20% | 40.20% | 39.39% | 84.82% | 87.20% | 9.50% |
| Claude 3.5 Haiku | 77.60% | 65.00% | 69.20% | 37.05% | 41.60% | **85.60%**| 88.10% | 8.02% |
| Cohere Aya-Vision 32B | 72.14% | 47.16% | 41.98% | 34.38% | 33.84% | 70.43% | 62.20% | 7.65% |
#### Vision
| Model | MMMU | MMMU PRO | Mathvista | ChartQA | DocVQA | AI2D | MM MT Bench |
|--------------------------------|------------|-----------|-----------|-----------|-----------|-------------|-------------|
| **Small 3.1 24B Instruct** | 64.00% | **49.25%**| **68.91%**| 86.24% | **94.08%**| **93.72%** | **7.3** |
| Gemma 3 27B IT | **64.90%** | 48.38% | 67.60% | 76.00% | 86.60% | 84.50% | 7 |
| GPT4o Mini | 59.40% | 37.60% | 56.70% | 76.80% | 86.70% | 88.10% | 6.6 |
| Claude 3.5 Haiku | 60.50% | 45.03% | 61.60% | **87.20%**| 90.00% | 92.10% | 6.5 |
| Cohere Aya-Vision 32B | 48.20% | 31.50% | 50.10% | 63.04% | 72.40% | 82.57% | 4.1 |
### Multilingual Evals
| Model | Average | European | East Asian | Middle Eastern |
|--------------------------------|------------|------------|------------|----------------|
| **Small 3.1 24B Instruct** | **71.18%** | **75.30%** | **69.17%** | 69.08% |
| Gemma 3 27B IT | 70.19% | 74.14% | 65.65% | 70.76% |
| GPT4o Mini | 70.36% | 74.21% | 65.96% | **70.90%** |
| Claude 3.5 Haiku | 70.16% | 73.45% | 67.05% | 70.00% |
| Cohere Aya-Vision 32B | 62.15% | 64.70% | 57.61% | 64.12% |
### Long Context Evals
| Model | LongBench v2 | RULER 32K | RULER 128K |
|--------------------------------|-----------------|-------------|------------|
| **Small 3.1 24B Instruct** | **37.18%** | **93.96%** | 81.20% |
| Gemma 3 27B IT | 34.59% | 91.10% | 66.00% |
| GPT4o Mini | 29.30% | 90.20% | 65.8% |
| Claude 3.5 Haiku | 35.19% | 92.60% | **91.90%** |
## Basic Instruct Template (V7-Tekken)
```
<s>[SYSTEM_PROMPT]<system prompt>[/SYSTEM_PROMPT][INST]<user message>[/INST]<assistant response></s>[INST]<user message>[/INST]
```
*`<system_prompt>`, `<user message>` and `<assistant response>` are placeholders.*
***Please make sure to use [mistral-common](https://github.com/mistralai/mistral-common) as the source of truth***
## Usage
The model can be used with the following frameworks;
- [`vllm (recommended)`](https://github.com/vllm-project/vllm): See [here](#vllm)
**Note 1**: We recommend using a relatively low temperature, such as `temperature=0.15`.
**Note 2**: Make sure to add a system prompt to the model to best tailer it for your needs. If you want to use the model as a general assistant, we recommend the following
system prompt:
```
system_prompt = """You are Mistral Small 3.1, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.
You power an AI assistant called Le Chat.
Your knowledge base was last updated on 2023-10-01.
The current date is {today}.
When you're not sure about some information, you say that you don't have the information and don't make up anything.
If the user's question is not clear, ambiguous, or does not provide enough context for you to accurately answer the question, you do not try to answer it right away and you rather ask the user to clarify their request (e.g. "What are some good restaurants around me?" => "Where are you?" or "When is the next flight to Tokyo" => "Where do you travel from?").
You are always very attentive to dates, in particular you try to resolve dates (e.g. "yesterday" is {yesterday}) and when asked about information at specific dates, you discard information that is at another date.
You follow these instructions in all languages, and always respond to the user in the language they use or request.
Next sections describe the capabilities that you have.
# WEB BROWSING INSTRUCTIONS
You cannot perform any web search or access internet to open URLs, links etc. If it seems like the user is expecting you to do so, you clarify the situation and ask the user to copy paste the text directly in the chat.
# MULTI-MODAL INSTRUCTIONS
You have the ability to read images, but you cannot generate images. You also cannot transcribe audio files or videos.
You cannot read nor transcribe audio files or videos."""
```
### vLLM (recommended)
We recommend using this model with the [vLLM library](https://github.com/vllm-project/vllm)
to implement production-ready inference pipelines.
**_Installation_**
Make sure you install [`vLLM >= 0.8.1`](https://github.com/vllm-project/vllm/releases/tag/v0.8.1):
```
pip install vllm --upgrade
```
Doing so should automatically install [`mistral_common >= 1.5.4`](https://github.com/mistralai/mistral-common/releases/tag/v1.5.4).
To check:
```
python -c "import mistral_common; print(mistral_common.__version__)"
```
You can also make use of a ready-to-go [docker image](https://github.com/vllm-project/vllm/blob/main/Dockerfile) or on the [docker hub](https://hub.docker.com/layers/vllm/vllm-openai/latest/images/sha256-de9032a92ffea7b5c007dad80b38fd44aac11eddc31c435f8e52f3b7404bbf39).
#### Server
We recommand that you use Mistral-Small-3.1-24B-Instruct-2503 in a server/client setting.
1. Spin up a server:
```
vllm serve mistralai/Mistral-Small-3.1-24B-Instruct-2503 --tokenizer_mode mistral --config_format mistral --load_format mistral --tool-call-parser mistral --enable-auto-tool-choice --limit_mm_per_prompt 'image=10' --tensor-parallel-size 2
```
**Note:** Running Mistral-Small-3.1-24B-Instruct-2503 on GPU requires ~55 GB of GPU RAM in bf16 or fp16.
2. To ping the client you can use a simple Python snippet.
```py
import requests
import json
from huggingface_hub import hf_hub_download
from datetime import datetime, timedelta
url = "http://<your-server-url>:8000/v1/chat/completions"
headers = {"Content-Type": "application/json", "Authorization": "Bearer token"}
model = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
def load_system_prompt(repo_id: str, filename: str) -> str:
file_path = hf_hub_download(repo_id=repo_id, filename=filename)
with open(file_path, "r") as file:
system_prompt = file.read()
today = datetime.today().strftime("%Y-%m-%d")
yesterday = (datetime.today() - timedelta(days=1)).strftime("%Y-%m-%d")
model_name = repo_id.split("/")[-1]
return system_prompt.format(name=model_name, today=today, yesterday=yesterday)
SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
image_url = "https://huggingface.co/datasets/patrickvonplaten/random_img/resolve/main/europe.png"
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{
"role": "user",
"content": [
{
"type": "text",
"text": "Which of the depicted countries has the best food? Which the second and third and fourth? Name the country, its color on the map and one its city that is visible on the map, but is not the capital. Make absolutely sure to only name a city that can be seen on the map.",
},
{"type": "image_url", "image_url": {"url": image_url}},
],
},
]
data = {"model": model, "messages": messages, "temperature": 0.15}
response = requests.post(url, headers=headers, data=json.dumps(data))
print(response.json()["choices"][0]["message"]["content"])
# Determining the "best" food is highly subjective and depends on personal preferences. However, based on general popularity and recognition, here are some countries known for their cuisine:
# 1. **Italy** - Color: Light Green - City: Milan
# - Italian cuisine is renowned worldwide for its pasta, pizza, and various regional specialties.
# 2. **France** - Color: Brown - City: Lyon
# - French cuisine is celebrated for its sophistication, including dishes like coq au vin, bouillabaisse, and pastries like croissants and éclairs.
# 3. **Spain** - Color: Yellow - City: Bilbao
# - Spanish cuisine offers a variety of flavors, from paella and tapas to jamón ibérico and churros.
# 4. **Greece** - Not visible on the map
# - Greek cuisine is known for dishes like moussaka, souvlaki, and baklava. Unfortunately, Greece is not visible on the provided map, so I cannot name a city.
# Since Greece is not visible on the map, I'll replace it with another country known for its good food:
# 4. **Turkey** - Color: Light Green (east part of the map) - City: Istanbul
# - Turkish cuisine is diverse and includes dishes like kebabs, meze, and baklava.
```
### Function calling
Mistral-Small-3.1-24-Instruct-2503 is excellent at function / tool calling tasks via vLLM. *E.g.:*
<details>
<summary>Example</summary>
```py
import requests
import json
from huggingface_hub import hf_hub_download
from datetime import datetime, timedelta
url = "http://<your-url>:8000/v1/chat/completions"
headers = {"Content-Type": "application/json", "Authorization": "Bearer token"}
model = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
def load_system_prompt(repo_id: str, filename: str) -> str:
file_path = hf_hub_download(repo_id=repo_id, filename=filename)
with open(file_path, "r") as file:
system_prompt = file.read()
today = datetime.today().strftime("%Y-%m-%d")
yesterday = (datetime.today() - timedelta(days=1)).strftime("%Y-%m-%d")
model_name = repo_id.split("/")[-1]
return system_prompt.format(name=model_name, today=today, yesterday=yesterday)
SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather in a given location",
"parameters": {
"type": "object",
"properties": {
"city": {
"type": "string",
"description": "The city to find the weather for, e.g. 'San Francisco'",
},
"state": {
"type": "string",
"description": "The state abbreviation, e.g. 'CA' for California",
},
"unit": {
"type": "string",
"description": "The unit for temperature",
"enum": ["celsius", "fahrenheit"],
},
},
"required": ["city", "state", "unit"],
},
},
},
{
"type": "function",
"function": {
"name": "rewrite",
"description": "Rewrite a given text for improved clarity",
"parameters": {
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "The input text to rewrite",
}
},
},
},
},
]
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{
"role": "user",
"content": "Could you please make the below article more concise?\n\nOpenAI is an artificial intelligence research laboratory consisting of the non-profit OpenAI Incorporated and its for-profit subsidiary corporation OpenAI Limited Partnership.",
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "bbc5b7ede",
"type": "function",
"function": {
"name": "rewrite",
"arguments": '{"text": "OpenAI is an artificial intelligence research laboratory consisting of the non-profit OpenAI Incorporated and its for-profit subsidiary corporation OpenAI Limited Partnership."}',
},
}
],
},
{
"role": "tool",
"content": '{"action":"rewrite","outcome":"OpenAI is a FOR-profit company."}',
"tool_call_id": "bbc5b7ede",
"name": "rewrite",
},
{
"role": "assistant",
"content": "---\n\nOpenAI is a FOR-profit company.",
},
{
"role": "user",
"content": "Can you tell me what the temperature will be in Dallas, in Fahrenheit?",
},
]
data = {"model": model, "messages": messages, "tools": tools, "temperature": 0.15}
response = requests.post(url, headers=headers, data=json.dumps(data))
print(response.json()["choices"][0]["message"]["tool_calls"])
# [{'id': '8PdihwL6d', 'type': 'function', 'function': {'name': 'get_current_weather', 'arguments': '{"city": "Dallas", "state": "TX", "unit": "fahrenheit"}'}}]
```
</details>
#### Offline
```py
from vllm import LLM
from vllm.sampling_params import SamplingParams
from datetime import datetime, timedelta
SYSTEM_PROMPT = "You are a conversational agent that always answers straight to the point, always end your accurate response with an ASCII drawing of a cat."
user_prompt = "Give me 5 non-formal ways to say 'See you later' in French."
messages = [
{
"role": "system",
"content": SYSTEM_PROMPT
},
{
"role": "user",
"content": user_prompt
},
]
model_name = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
# note that running this model on GPU requires over 60 GB of GPU RAM
llm = LLM(model=model_name, tokenizer_mode="mistral")
sampling_params = SamplingParams(max_tokens=512, temperature=0.15)
outputs = llm.chat(messages, sampling_params=sampling_params)
print(outputs[0].outputs[0].text)
# Here are five non-formal ways to say "See you later" in French:
# 1. **À plus tard** - Until later
# 2. **À toute** - See you soon (informal)
# 3. **Salut** - Bye (can also mean hi)
# 4. **À plus** - See you later (informal)
# 5. **Ciao** - Bye (informal, borrowed from Italian)
# ```
# /\_/\
# ( o.o )
# > ^ <
# ```
```
### Transformers (untested)
Transformers-compatible model weights are also uploaded (thanks a lot @cyrilvallez).
However the transformers implementation was **not throughly tested**, but only on "vibe-checks".
Hence, we can only ensure 100% correct behavior when using the original weight format with vllm (see above). |
Mungert/granite-3.2-2b-instruct-GGUF | Mungert | 2025-06-15T19:40:53Z | 658 | 6 | transformers | [
"transformers",
"gguf",
"language",
"granite-3.2",
"text-generation",
"arxiv:0000.00000",
"base_model:ibm-granite/granite-3.1-2b-instruct",
"base_model:quantized:ibm-granite/granite-3.1-2b-instruct",
"license:apache-2.0",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-18T21:59:43Z | ---
pipeline_tag: text-generation
inference: false
license: apache-2.0
library_name: transformers
tags:
- language
- granite-3.2
base_model:
- ibm-granite/granite-3.1-2b-instruct
---
# <span style="color: #7FFF7F;">granite-3.2-2b-instruct GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `granite-3.2-2b-instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `granite-3.2-2b-instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `granite-3.2-2b-instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `granite-3.2-2b-instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `granite-3.2-2b-instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `granite-3.2-2b-instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `granite-3.2-2b-instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `granite-3.2-2b-instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `granite-3.2-2b-instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `granite-3.2-2b-instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `granite-3.2-2b-instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# Granite-3.2-2B-Instruct
**Model Summary:**
Granite-3.2-2B-Instruct is an 2-billion-parameter, long-context AI model fine-tuned for thinking capabilities. Built on top of [Granite-3.1-2B-Instruct](https://huggingface.co/ibm-granite/granite-3.1-2b-instruct), it has been trained using a mix of permissively licensed open-source datasets and internally generated synthetic data designed for reasoning tasks. The model allows controllability of its thinking capability, ensuring it is applied only when required.
- **Developers:** Granite Team, IBM
- **Website**: [Granite Docs](https://www.ibm.com/granite/docs/)
- **Release Date**: February 26th, 2025
- **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
**Supported Languages:**
English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. However, users may finetune this Granite model for languages beyond these 12 languages.
**Intended Use:**
This model is designed to handle general instruction-following tasks and can be integrated into AI assistants across various domains, including business applications.
**Capabilities**
* **Thinking**
* Summarization
* Text classification
* Text extraction
* Question-answering
* Retrieval Augmented Generation (RAG)
* Code related tasks
* Function-calling tasks
* Multilingual dialog use cases
* Long-context tasks including long document/meeting summarization, long document QA, etc.
**Generation:**
This is a simple example of how to use Granite-3.2-2B-Instruct model.
Install the following libraries:
```shell
pip install torch torchvision torchaudio
pip install accelerate
pip install transformers
```
Then, copy the snippet from the section that is relevant for your use case.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
import torch
model_path="ibm-granite/granite-3.2-2b-instruct"
device="cuda"
model = AutoModelForCausalLM.from_pretrained(
model_path,
device_map=device,
torch_dtype=torch.bfloat16,
)
tokenizer = AutoTokenizer.from_pretrained(
model_path
)
conv = [{"role": "user", "content":"You have 10 liters of a 30% acid solution. How many liters of a 70% acid solution must be added to achieve a 50% acid mixture?"}]
input_ids = tokenizer.apply_chat_template(conv, return_tensors="pt", thinking=True, return_dict=True, add_generation_prompt=True).to(device)
set_seed(42)
output = model.generate(
**input_ids,
max_new_tokens=8192,
)
prediction = tokenizer.decode(output[0, input_ids["input_ids"].shape[1]:], skip_special_tokens=True)
print(prediction)
```
**Example Outputs**
- thinking=True
```md
Here is my thought process:
1. First, let's understand the composition of the initial 10 liters of 30% acid solution:
- Acid content = 30% of 10 liters = 0.30 * 10 = 3 liters
2. We want to end up with a 50% acid solution. Let's denote the volume of the 70% acid solution to be added as 'x' liters.
3. The amount of acid in the final mixture (50% solution) should equal the initial acid content plus the acid from the added 70% solution.
- Total acid in final mixture = 0.50 * (10 + x)
4. The equation representing the total acid content is:
3 (from initial solution) + 0.70 * x (from added solution) = 0.50 * (10 + x)
5. Now, we solve for 'x':
3 + 0.70x = 5 + 0.50x
0.70x - 0.50x = 5 - 3
0.20x = 2
x = 2 / 0.20
x = 10 liters
6. Therefore, 10 liters of a 70% acid solution must be added to the initial 10 liters of 30% acid solution to achieve a 50% acid mixture.
Here is my response:
To solve this problem, we need to determine how much of a 70% acid solution must be added to a 10-liter solution that is already 30% acid to create a new mixture that is 50% acid. Here's how to calculate it:
1. **Initial Acid Content**:
- The initial 10-liter solution contains 3 liters of acid (since 30% of 10 liters).
2. **Desired Final Acid Content**:
- We want the final mixture to be 50% acid.
3. **Equation Setup**:
- Let 'x' be the volume (in liters) of the 70% acid solution to be added.
- The equation representing the total acid content in the final mixture (which is 10 liters from the initial solution plus 'x' liters of the added solution) should equal 50% of the total volume (10 + x).
- This gives us: 3 (acid from initial solution) + 0.70x (acid from added solution) = 0.50(10 + x)
4. **Solving the Equation**:
- First, simplify the right side of the equation: 0.50(10 + x) = 5 + 0.50x
- Now, the equation is: 3 + 0.70x = 5 + 0.50x
- Subtract 0.50x from both sides: 0.20x = 2
- Divide both sides by 0.20: x = 2 / 0.20
- x = 10 liters
5. **Conclusion**:
- To achieve a 50% acid mixture from the initial 30% solution, you need to add 10 liters of a 70% acid solution.
So, the final answer is that 10 liters of a 70% acid solution must be added to the initial 10 liters of 30% acid solution.
```
- thinking=False
```md
To solve this problem, we'll use the concept of volume and concentration. Let's denote the amount of 70% acid solution we need to add as "x" liters.
First, let's find out how much acid is in the initial 10-liter solution:
Acid in initial solution = 30% of 10 liters = 0.30 * 10 = 3 liters
Now, we want to end up with a 50% acid solution in a total volume of (10 + x) liters. Let's denote the final volume as V.
Final acid concentration = 50%
Final acid amount = 50% of V = 0.50 * V
We know the initial acid amount and the final acid amount, so we can set up an equation:
Initial acid amount + Acid from added solution = Final acid amount
3 liters + (70% of x) = 0.50 * (10 + x)
Now, let's solve for x:
0.70x + 3 = 0.50 * 10 + 0.50x
0.70x - 0.50x = 0.50 * 10 - 3
0.20x = 5 - 3
0.20x = 2
x = 2 / 0.20
x = 10 liters
So, you need to add 10 liters of a 70% acid solution to the initial 10-liter 30% acid solution to achieve a 50% acid mixture.
```
**Evaluation Results:**
<table>
<thead>
<tr>
<th style="text-align:left; background-color: #001d6c; color: white;">Models</th>
<th style="text-align:center; background-color: #001d6c; color: white;">ArenaHard</th>
<th style="text-align:center; background-color: #001d6c; color: white;">Alpaca-Eval-2</th>
<th style="text-align:center; background-color: #001d6c; color: white;">MMLU</th>
<th style="text-align:center; background-color: #001d6c; color: white;">PopQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">TruthfulQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">BigBenchHard</th>
<th style="text-align:center; background-color: #001d6c; color: white;">DROP</th>
<th style="text-align:center; background-color: #001d6c; color: white;">GSM8K</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval+</th>
<th style="text-align:center; background-color: #001d6c; color: white;">IFEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">AttaQ</th>
</tr></thead>
<tbody>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Llama-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">36.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">27.22</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">69.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">52.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">72.66</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.48</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.24</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.32</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">80.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.10</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.43</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">DeepSeek-R1-Distill-Llama-8B</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">17.17</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">21.85</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">45.80</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">13.25</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">47.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.71</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">44.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">72.18</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.54</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">62.91</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.50</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.87</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Qwen-2.5-7B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">25.44</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">30.34</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.30</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">18.12</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">63.06</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">70.40</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">54.71</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">84.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">93.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.91</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.90</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">81.90</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">DeepSeek-R1-Distill-Qwen-7B</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">10.36</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">15.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.72</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">9.94</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">47.14</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.04</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.76</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.47</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.89</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.43</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.10</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">42.45</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">37.58</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">30.34</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.77</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.7</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">65.84</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">68.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.78</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.15</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.63</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">73.20</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.73</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.1-2B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">23.3</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">27.17</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">57.11</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">20.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">54.46</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">18.68</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">79.45</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">75.26</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">63.59</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">84.7</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;">Granite-3.2-8B-Instruct</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">55.25</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.19</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.79</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">28.04</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">66.92</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">64.77</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">50.95</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">81.65</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">89.35</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.72</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">74.31</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">85.42</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.2-2B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">24.86</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">34.51</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">57.18</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">20.56</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">59.8</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">52.27</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">21.12</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">67.02</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">80.13</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">73.39</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">61.55</td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">83.23</td>
</tr>
</tbody></table>
**Training Data:**
Overall, our training data is largely comprised of two key sources: (1) publicly available datasets with permissive license, (2) internal synthetically generated data targeted to enhance reasoning capabilites.
<!-- A detailed attribution of datasets can be found in [Granite 3.2 Technical Report (coming soon)](#), and [Accompanying Author List](https://github.com/ibm-granite/granite-3.0-language-models/blob/main/author-ack.pdf). -->
**Infrastructure:**
We train Granite-3.2-2B-Instruct using IBM's super computing cluster, Blue Vela, which is outfitted with NVIDIA H100 GPUs. This cluster provides a scalable and efficient infrastructure for training our models over thousands of GPUs.
**Ethical Considerations and Limitations:**
Granite-3.2-2B-Instruct builds upon Granite-3.1-2B-Instruct, leveraging both permissively licensed open-source and select proprietary data for enhanced performance. Since it inherits its foundation from the previous model, all ethical considerations and limitations applicable to [Granite-3.1-2B-Instruct](https://huggingface.co/ibm-granite/granite-3.1-2b-instruct) remain relevant.
**Resources**
- ⭐️ Learn about the latest updates with Granite: https://www.ibm.com/granite
- 📄 Get started with tutorials, best practices, and prompt engineering advice: https://www.ibm.com/granite/docs/
- 💡 Learn about the latest Granite learning resources: https://ibm.biz/granite-learning-resources
<!-- ## Citation
```
@misc{granite-models,
author = {author 1, author2, ...},
title = {},
journal = {},
volume = {},
year = {2024},
url = {https://arxiv.org/abs/0000.00000},
}
``` --> |
Mungert/rwkv7-0.4B-world-GGUF | Mungert | 2025-06-15T19:40:45Z | 517 | 2 | null | [
"gguf",
"text-generation",
"en",
"zh",
"ja",
"ko",
"fr",
"ar",
"es",
"pt",
"base_model:BlinkDL/rwkv-7-world",
"base_model:quantized:BlinkDL/rwkv-7-world",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-18T09:56:18Z | ---
license: apache-2.0
language:
- en
- zh
- ja
- ko
- fr
- ar
- es
- pt
metrics:
- accuracy
base_model:
- BlinkDL/rwkv-7-world
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">rwkv7-0.4B-world GGUF Models</span>
Note: you must use latest llama.cpp https://github.com/ggml-org/llama.cpp to run this model with llama.cpp
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `rwkv7-0.4B-world-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `rwkv7-0.4B-world-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `rwkv7-0.4B-world-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `rwkv7-0.4B-world-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `rwkv7-0.4B-world-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `rwkv7-0.4B-world-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `rwkv7-0.4B-world-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `rwkv7-0.4B-world-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `rwkv7-0.4B-world-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `rwkv7-0.4B-world-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `rwkv7-0.4B-world-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# rwkv7-0.4B-world
<!-- Provide a quick summary of what the model is/does. -->
This is RWKV-7 model under flash-linear attention format.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Bo Peng, Yu Zhang, Songlin Yang, Ruichong Zhang
- **Funded by:** RWKV Project (Under LF AI & Data Foundation)
- **Model type:** RWKV7
- **Language(s) (NLP):** English
- **License:** Apache-2.0
- **Parameter count:** 0.450B
- **Tokenizer:** RWKV World tokenizer
- **Vocabulary size:** 65,536
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** https://github.com/fla-org/flash-linear-attention ; https://github.com/BlinkDL/RWKV-LM
- **Paper:** With in Progress
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
Install `flash-linear-attention` and the latest version of `transformers` before using this model:
```bash
pip install git+https://github.com/fla-org/flash-linear-attention
pip install 'transformers>=4.48.0'
```
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
You can use this model just as any other HuggingFace models:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('fla-hub/rwkv7-0.4B-world', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('fla-hub/rwkv7-0.4B-world', trust_remote_code=True)
model = model.cuda()
prompt = "What is a large language model?"
messages = [
{"role": "user", "content": "Who are you?"},
{"role": "assistant", "content": "I am a GPT-3 based model."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=1024,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)[0]
print(response)
```
## Training Details
### Training Data
This model is trained on the World v3 with a total of 3.119 trillion tokens.
#### Training Hyperparameters
- **Training regime:** bfloat16, lr 4e-4 to 1e-5 "delayed" cosine decay, wd 0.1 (with increasing batch sizes during the middle)
## FAQ
Q: safetensors metadata is none.
A: upgrade transformers to >=4.48.0: `pip install 'transformers>=4.48.0'` |
Mungert/rwkv7-1.5B-world-GGUF | Mungert | 2025-06-15T19:40:40Z | 811 | 1 | null | [
"gguf",
"text-generation",
"en",
"zh",
"ja",
"ko",
"fr",
"ar",
"es",
"pt",
"base_model:BlinkDL/rwkv-7-world",
"base_model:quantized:BlinkDL/rwkv-7-world",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-18T08:04:12Z | ---
license: apache-2.0
language:
- en
- zh
- ja
- ko
- fr
- ar
- es
- pt
metrics:
- accuracy
base_model:
- BlinkDL/rwkv-7-world
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">rwkv7-1.5B-world GGUF Models</span>
Note: you must use latest llama.cpp https://github.com/ggml-org/llama.cpp to run this model with llama.cpp
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `rwkv7-1.5B-world-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `rwkv7-1.5B-world-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `rwkv7-1.5B-world-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `rwkv7-1.5B-world-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `rwkv7-1.5B-world-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `rwkv7-1.5B-world-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `rwkv7-1.5B-world-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `rwkv7-1.5B-world-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `rwkv7-1.5B-world-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `rwkv7-1.5B-world-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `rwkv7-1.5B-world-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# rwkv7-1.5B-world
<!-- Provide a quick summary of what the model is/does. -->
This is RWKV-7 model under flash-linear attention format.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Bo Peng, Yu Zhang, Songlin Yang, Ruichong Zhang
- **Funded by:** RWKV Project (Under LF AI & Data Foundation)
- **Model type:** RWKV7
- **Language(s) (NLP):** English
- **License:** Apache-2.0
- **Parameter count:** 1.52B
- **Tokenizer:** RWKV World tokenizer
- **Vocabulary size:** 65,536
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** https://github.com/fla-org/flash-linear-attention ; https://github.com/BlinkDL/RWKV-LM
- **Paper:** With in Progress
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
Install `flash-linear-attention` and the latest version of `transformers` before using this model:
```bash
pip install git+https://github.com/fla-org/flash-linear-attention
pip install 'transformers>=4.48.0'
```
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
You can use this model just as any other HuggingFace models:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('fla-hub/rwkv7-1.5B-world', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('fla-hub/rwkv7-1.5B-world', trust_remote_code=True)
model = model.cuda()
prompt = "What is a large language model?"
messages = [
{"role": "user", "content": "Who are you?"},
{"role": "assistant", "content": "I am a GPT-3 based model."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=1024,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)[0]
print(response)
```
## Training Details
### Training Data
This model is trained on the World v3 with a total of 3.119 trillion tokens.
#### Training Hyperparameters
- **Training regime:** bfloat16, lr 4e-4 to 1e-5 "delayed" cosine decay, wd 0.1 (with increasing batch sizes during the middle)
- **Final Loss:** 1.9965
- **Token Count:** 3.119 trillion
## Evaluation
#### Metrics
`lambada_openai`:
before conversion: ppl 4.13 acc 69.4%
after conversion: ppl 4.26 acc 68.8% (without apply temple)
## FAQ
Q: safetensors metadata is none.
A: upgrade transformers to >=4.48.0: `pip install 'transformers>=4.48.0'` |
Mungert/DeepHermes-3-Llama-3-8B-Preview-GGUF | Mungert | 2025-06-15T19:40:29Z | 1,885 | 5 | transformers | [
"transformers",
"gguf",
"Llama-3",
"instruct",
"finetune",
"chatml",
"gpt4",
"synthetic data",
"distillation",
"function calling",
"json mode",
"axolotl",
"roleplaying",
"chat",
"reasoning",
"r1",
"vllm",
"en",
"base_model:meta-llama/Llama-3.1-8B",
"base_model:quantized:meta-llama/Llama-3.1-8B",
"license:llama3",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-03-17T14:30:06Z | ---
language:
- en
license: llama3
tags:
- Llama-3
- instruct
- finetune
- chatml
- gpt4
- synthetic data
- distillation
- function calling
- json mode
- axolotl
- roleplaying
- chat
- reasoning
- r1
- vllm
base_model: meta-llama/Meta-Llama-3.1-8B
widget:
- example_title: Hermes 3
messages:
- role: system
content: >-
You are a sentient, superintelligent artificial general intelligence, here
to teach and assist me.
- role: user
content: What is the meaning of life?
model-index:
- name: DeepHermes-3-Llama-3.1-8B
results: []
library_name: transformers
---
# <span style="color: #7FFF7F;">DeepHermes-3-Llama-3-8B-Preview GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `DeepHermes-3-Llama-3-8B-Preview-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `DeepHermes-3-Llama-3-8B-Preview-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `DeepHermes-3-Llama-3-8B-Preview-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `DeepHermes-3-Llama-3-8B-Preview-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `DeepHermes-3-Llama-3-8B-Preview-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `DeepHermes-3-Llama-3-8B-Preview-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `DeepHermes-3-Llama-3-8B-Preview-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `DeepHermes-3-Llama-3-8B-Preview-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `DeepHermes-3-Llama-3-8B-Preview-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `DeepHermes-3-Llama-3-8B-Preview-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `DeepHermes-3-Llama-3-8B-Preview-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# DeepHermes 3 - Llama-3.1 8B
## Model Description
DeepHermes 3 Preview is the latest version of our flagship Hermes series of LLMs by Nous Research, and one of the first models in the world to unify Reasoning (long chains of thought that improve answer accuracy) and normal LLM response modes into one model. We have also improved LLM annotation, judgement, and function calling.
DeepHermes 3 Preview is one of the first LLM models to unify both "intuitive", traditional mode responses and **long chain of thought reasoning** responses into a single model, toggled by a system prompt.
Hermes 3, the predecessor of DeepHermes 3, is a generalist language model with many improvements over Hermes 2, including advanced agentic capabilities, much better roleplaying, reasoning, multi-turn conversation, long context coherence, and improvements across the board.
The ethos of the Hermes series of models is focused on aligning LLMs to the user, with powerful steering capabilities and control given to the end user.
*This is a preview Hermes with early reasoning capabilities, distilled from R1 across a variety of tasks that benefit from reasoning and objectivity. Some quirks may be discovered! Please let us know any interesting findings or issues you discover!*
## Note: To toggle REASONING ON, you must use the following system prompt:
```
You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem.
```
# Nous API
This model is also available on our new API product - Check out the API and sign up for the waitlist here:
https://portal.nousresearch.com/
# Example Outputs:
# Benchmarks
## Benchmarks for **Reasoning Mode** on vs off:
*Reasoning ON benchmarks aquired by running HuggingFace's open-r1 reasoning mode evaluation suite, and scores for reasoning mode OFF aquired by running LM-Eval-Harness Benchmark Suite*
*Upper bound determined by measuring the % gained over Hermes 3 3 & 70b by MATH_VERIFY compared to eleuther eval harness, which ranged betweeen 33% and 50% gain in MATH Hard benchmark on retested models by them compared to eval harness reported scores*
## Benchmarks in **Non-Reasoning Mode** against Llama-3.1-8B-Instruct
# Prompt Format
DeepHermes 3 now uses Llama-Chat format as the prompt format, opening up a more unified, structured system for engaging the LLM in multi-turn chat dialogue.
System prompts allow steerability and interesting new ways to interact with an LLM, guiding rules, roles, and stylistic choices of the model.
## Deep Thinking Mode - Deep Hermes Preview can activate long chain of thought with a system prompt.
```
You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem.
```
For an example of using deep reasoning mode with HuggingFace Transformers:
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
import flash_attn
import time
tokenizer = AutoTokenizer.from_pretrained("NousResearch/DeepHermes-3-Llama-3-8B-Preview")
model = AutoModelForCausalLM.from_pretrained(
"NousResearch/DeepHermes-3-Llama-3-8B-Preview",
torch_dtype=torch.float16,
device_map="auto",
attn_implementation="flash_attention_2",
)
messages = [
{
"role": "system",
"content": "You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem."
},
{
"role": "user",
"content": "What is y if y=2*2-4+(3*2)"
}
]
input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors='pt').to("cuda")
generated_ids = model.generate(input_ids, max_new_tokens=2500, temperature=0.8, repetition_penalty=1.1, do_sample=True, eos_token_id=tokenizer.eos_token_id)
print(f"Generated Tokens: {generated_ids.shape[-1:]}")
response = tokenizer.decode(generated_ids[0], skip_special_tokens=True, clean_up_tokenization_space=True)
print(f"Response: {response}")
```
Please note, for difficult problems DeepHermes can think using as many as 13,000 tokens. You may need to increase `max_new_tokens` to be much larger than 2500 for difficult problems.
## Standard "Intuitive" Response Mode
Prompt with system instruction (Use whatever system prompt you like, this is just an example!):
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
import flash_attn
import time
tokenizer = AutoTokenizer.from_pretrained("NousResearch/DeepHermes-3-Llama-3-8B-Preview")
model = AutoModelForCausalLM.from_pretrained(
"NousResearch/DeepHermes-3-Llama-3-8B-Preview",
torch_dtype=torch.float16,
device_map="auto",
attn_implementation="flash_attention_2",
)
messages = [
{
"role": "system",
"content": "You are Hermes, an AI assistant"
},
{
"role": "user",
"content": "What are the most interesting things to do in Paris?"
}
]
input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors='pt').to("cuda")
generated_ids = model.generate(input_ids, max_new_tokens=2500, temperature=0.8, repetition_penalty=1.1, do_sample=True, eos_token_id=tokenizer.eos_token_id)
print(f"Generated Tokens: {generated_ids.shape[-1:]}")
response = tokenizer.decode(generated_ids[0], skip_special_tokens=True, clean_up_tokenization_space=True)
print(f"Response: {response}")
```
## VLLM Inference
You can also run this model with vLLM, by running the following in your terminal after `pip install vllm`
`vllm serve NousResearch/DeepHermes-3-Llama-3-8B-Preview`
You may then use the model over API using the OpenAI library just like you would call OpenAI's API.
## Prompt Format for Function Calling
Our model was trained on specific system prompts and structures for Function Calling.
You should use the system role with this message, followed by a function signature json as this example shows here.
```
<|start_header_id|>system<|end_header_id|>
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools: <tools> {"type": "function", "function": {"name": "get_stock_fundamentals", "description": "get_stock_fundamentals(symbol: str) -> dict - Get fundamental data for a given stock symbol using yfinance API.\\n\\n Args:\\n symbol (str): The stock symbol.\\n\\n Returns:\\n dict: A dictionary containing fundamental data.\\n Keys:\\n - \'symbol\': The stock symbol.\\n - \'company_name\': The long name of the company.\\n - \'sector\': The sector to which the company belongs.\\n - \'industry\': The industry to which the company belongs.\\n - \'market_cap\': The market capitalization of the company.\\n - \'pe_ratio\': The forward price-to-earnings ratio.\\n - \'pb_ratio\': The price-to-book ratio.\\n - \'dividend_yield\': The dividend yield.\\n - \'eps\': The trailing earnings per share.\\n - \'beta\': The beta value of the stock.\\n - \'52_week_high\': The 52-week high price of the stock.\\n - \'52_week_low\': The 52-week low price of the stock.", "parameters": {"type": "object", "properties": {"symbol": {"type": "string"}}, "required": ["symbol"]}}} </tools> Use the following pydantic model json schema for each tool call you will make: {"properties": {"arguments": {"title": "Arguments", "type": "object"}, "name": {"title": "Name", "type": "string"}}, "required": ["arguments", "name"], "title": "FunctionCall", "type": "object"} For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
<tool_call>
{"arguments": <args-dict>, "name": <function-name>}
</tool_call><|eot_id|><|start_header_id|>user<|end_header_id|>
```
To complete the function call, create a user prompt that follows the above system prompt, like so:
```
Fetch the stock fundamentals data for Tesla (TSLA)<|eot_id|><|start_header_id|>assistant<|end_header_id|>
```
The model will then generate a tool call, which your inference code must parse, and plug into a function (see example inference code here: https://github.com/NousResearch/Hermes-Function-Calling):
```
<tool_call>
{"arguments": {"symbol": "TSLA"}, "name": "get_stock_fundamentals"}
</tool_call><|eot_id|><|start_header_id|>tool<|end_header_id|>
```
Once you parse the tool call, call the api and get the returned values for the call, and pass it back in as a new role, `tool` like so:
```
<tool_response>
{"name": "get_stock_fundamentals", "content": {'symbol': 'TSLA', 'company_name': 'Tesla, Inc.', 'sector': 'Consumer Cyclical', 'industry': 'Auto Manufacturers', 'market_cap': 611384164352, 'pe_ratio': 49.604652, 'pb_ratio': 9.762013, 'dividend_yield': None, 'eps': 4.3, 'beta': 2.427, '52_week_high': 299.29, '52_week_low': 152.37}}
</tool_response>
<|eot_id|><|start_header_id|>assistant<|end_header_id|>
```
The assistant will then read in that data from the function's response, and generate a natural language response:
```
The stock fundamentals data for Tesla (TSLA) are as follows:
- **Symbol**: TSLA
- **Company Name**: Tesla, Inc.
- **Sector**: Consumer Cyclical
- **Industry**: Auto Manufacturers
- **Market Capitalization**: $566,160,130,480
- **Forward Price-to-Earnings Ratio (PE Ratio)**: 42.73
- **Price-to-Book Ratio (PB Ratio)**: 9.04
- **Dividend Yield**: N/A
- **Trailing Earnings Per Share (EPS)**: $4.3
- **Beta Value of the Stock**: 2.42
- **52-Week High Price of the Stock**: $299.29
- **52-Week Low Price of the Stock**: $152.37
This information provides a snapshot of Tesla's financial position and performance based on the fundamental data obtained from the yfinance API. It shows that Tesla has a substantial market capitalization and a relatively high P/E and P/B ratio compared to other stocks in its industry. The company does not pay a dividend at the moment, which is reflected by a 'Dividend Yield' of 'None'. The Beta value indicates that Tesla's stock has a moderate level of volatility relative to the market. The 52-week high and low prices give an idea of the stock's range over the past year. This data can be useful when assessing investment opportunities and making investment decisions.<|eot_id|><|start_header_id|>user<|end_header_id|>
```
## Prompt Format for JSON Mode / Structured Outputs
Our model was also trained on a specific system prompt for Structured Outputs, which should respond with **only** a json object response, in a specific json schema.
Your schema can be made from a pydantic object using our codebase, with the standalone script `jsonmode.py` available here: https://github.com/NousResearch/Hermes-Function-Calling/tree/main
```
<|start_header_id|>system<|end_header_id|>
You are a helpful assistant that answers in JSON. Here's the json schema you must adhere to:\n<schema>\n{schema}\n</schema><|eot_id|>
```
Given the {schema} that you provide, it should follow the format of that json to create its response, all you have to do is give a typical user prompt, and it will respond in JSON.
## Inference Code for Function Calling:
All code for utilizing, parsing, and building function calling templates is available on our github:
[https://github.com/NousResearch/Hermes-Function-Calling](https://github.com/NousResearch/Hermes-Function-Calling)
## Quantized Versions:
GGUF Quants: https://huggingface.co/NousResearch/DeepHermes-3-Llama-3-8B-Preview-GGUF
# How to cite:
```bibtext
@misc{
title={DeepHermes 3 Preview},
author={Teknium and Roger Jin and Chen Guang and Jai Suphavadeeprasit and Jeffrey Quesnelle},
year={2025}
}
``` |
Mungert/TriLM_830M_Unpacked-GGUF | Mungert | 2025-06-15T19:40:14Z | 219 | 0 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | 2025-03-17T01:50:08Z | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_830M_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_830M_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_830M_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_830M_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_830M_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_830M_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_830M_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_830M_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_830M_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_830M_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_830M_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_830M_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# TriLM 830M Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_830M_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/TriLM_560M_Unpacked-GGUF | Mungert | 2025-06-15T19:40:11Z | 186 | 1 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | 2025-03-17T00:51:35Z | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_560M_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_560M_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_560M_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_560M_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_560M_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_560M_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_560M_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_560M_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_560M_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_560M_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_560M_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_560M_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# TriLM 560M Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_560M_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/Refact-1_6B-fim-GGUF | Mungert | 2025-06-15T19:40:09Z | 353 | 3 | transformers | [
"transformers",
"gguf",
"code",
"text-generation",
"en",
"dataset:bigcode/the-stack-dedup",
"dataset:rombodawg/2XUNCENSORED_MegaCodeTraining188k",
"dataset:bigcode/commitpackft",
"arxiv:2108.12409",
"arxiv:1607.06450",
"arxiv:1910.07467",
"arxiv:1911.02150",
"license:bigscience-openrail-m",
"model-index",
"endpoints_compatible",
"region:us",
"imatrix"
] | text-generation | 2025-03-17T00:36:19Z | ---
pipeline_tag: text-generation
inference: true
widget:
- text: 'def print_hello_world():'
example_title: Hello world
group: Python
license: bigscience-openrail-m
pretrain-datasets:
- books
- arxiv
- c4
- falcon-refinedweb
- wiki
- github-issues
- stack_markdown
- self-made dataset of permissive github code
datasets:
- bigcode/the-stack-dedup
- rombodawg/2XUNCENSORED_MegaCodeTraining188k
- bigcode/commitpackft
metrics:
- code_eval
library_name: transformers
tags:
- code
model-index:
- name: Refact-1.6B
results:
- task:
type: text-generation
dataset:
type: openai_humaneval
name: HumanEval
metrics:
- name: pass@1 (T=0.01)
type: pass@1
value: 32.0
verified: false
- name: pass@1 (T=0.2)
type: pass@1
value: 31.5
verified: false
- name: pass@10 (T=0.8)
type: pass@10
value: 53.0
verified: false
- name: pass@100 (T=0.8)
type: pass@100
value: 76.9
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize Python
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 35.8
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize JavaScript
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 31.6
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize Java
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 29.1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize Go
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize C++
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 26.3
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize Rust
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalSynthesize Average
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests Python
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 18.38
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests JavaScript
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 12.28
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests Java
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 15.12
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests Go
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests C++
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 13.17
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests Rust
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 2.8
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixTests Average
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs Python
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 26.92
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs JavaScript
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 26.85
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs Java
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 30.76
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs Go
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs C++
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 25.94
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs Rust
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 8.44
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalFixDocs Average
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain Python
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 26.46
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain JavaScript
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 17.86
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain Java
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 20.94
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain Go
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain C++
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 18.78
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain Rust
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: bigcode/humanevalpack
name: HumanEvalExplain Average
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: -1
verified: false
- task:
type: text-generation
dataset:
type: mbpp
name: MBPP
metrics:
- name: pass@1 (T=0.01)
type: pass@1
value: 31.15
verified: false
- task:
type: text-generation
dataset:
type: ds1000
name: DS-1000 (Overall Completion)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 10.1
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (C++)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 21.61
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (C#)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 13.91
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (D)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 9.5
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Go)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 53.57
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Java)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 21.58
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Julia)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 13.75
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (JavaScript)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 26.88
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Lua)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 15.26
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (PHP)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 23.04
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Perl)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 12.1
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Python)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 29.6
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (R)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 13.77
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Ruby)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 12.68
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Racket)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 4.29
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Rust)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 19.54
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Scala)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 18.33
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Bash)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 5.7
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (Swift)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 17.68
verified: false
- task:
type: text-generation
dataset:
type: nuprl/MultiPL-E
name: MultiPL-HumanEval (TypeScript)
metrics:
- name: pass@1 (T=0.2)
type: pass@1
value: 25
verified: false
language:
- en
---
# <span style="color: #7FFF7F;">Refact-1_6B-fim GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Refact-1_6B-fim-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Refact-1_6B-fim-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Refact-1_6B-fim-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Refact-1_6B-fim-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Refact-1_6B-fim-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Refact-1_6B-fim-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Refact-1_6B-fim-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Refact-1_6B-fim-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Refact-1_6B-fim-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Refact-1_6B-fim-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Refact-1_6B-fim-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# Refact-1.6B
Finally, the model we started training with our [blog post](https://refact.ai/blog/2023/applying-recent-innovations-to-train-model/) is ready 🎉
After fine-tuning on generated data, it beats Replit 3b, Stability Code 3b and many other models. It almost beats
StarCoder ten times the size!
Model | Size | HumanEval pass@1 | HumanEval pass@10 |
----------------------|---------------|--------------------|--------------------|
DeciCoder-1b | 1b | 19.1% | |
<b>Refact-1.6-fim</b> | <b>1.6b</b> | <b>32.0%</b> | <b>53.0%</b> |
StableCode | 3b | 20.2% | 33.8% |
ReplitCode v1 | 3b | 21.9% | |
CodeGen2.5-multi | 7b | 28.4% | 47.5% |
CodeLlama | 7b | 33.5% | 59.6% |
StarCoder | 15b | 33.6% | |
Likely, it's the best model for practical use in your IDE for code completion because it's smart and fast!
You can start using it right now by downloading the
[Refact plugin](https://refact.ai/). You can host the model yourself, too, using the
[open source docker container](https://github.com/smallcloudai/refact).
And it's multi-language (see MultiPL-HumanEval and other metrics below) and it works as a chat (see the section below).
# It Works As a Chat
The primary application of this model is code completion (infill) in multiple programming languages.
But it works as a chat quite well.
HumanEval results using instruction following (chat) format, against models specialized for chat only:
Model | Size | pass@1 | pass@10 |
-----------------------|--------|----------|----------|
<b>Refact-1.6-fim</b> | 1.6b | 38.4% | 55.6% |
StableCode-instruct | 3b | 26.9% | 36.2% |
OctoGeeX | 6b | 44.7% | |
CodeLlama-instruct | 7b | 34.8% | 64.3% |
CodeGen2.5-instruct | 7b | 36.2% | 60.87 |
CodeLlama-instruct | 13b | 42.7% | 71.6% |
StarChat-β | 15b | 33.5% | |
OctoCoder | 15b | 46.2% | |
# Example
Fill-in-the-middle uses special tokens to identify the prefix/middle/suffix part of the input and output:
```python
# pip install -q transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
checkpoint = "smallcloudai/Refact-1_6B-fim"
device = "cuda" # for GPU usage or "cpu" for CPU usage
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.from_pretrained(checkpoint, trust_remote_code=True).to(device)
prompt = '<fim_prefix>def print_hello_world():\n """<fim_suffix>\n print("Hello world!")<fim_middle>'
inputs = tokenizer.encode(prompt, return_tensors="pt").to(device)
outputs = model.generate(inputs, max_length=100, temperature=0.2)
print("-"*80)
print(tokenizer.decode(outputs[0]))
```
# Chat Format
The same model works as chat (experimental).
```python
prompt_template = "<empty_output>SYSTEM {system}\n" \
"<empty_output>USER {query}\n" \
"<empty_output>ASSISTANT"
prompt = prompt_template.format(system="You are a programming assistant",
query="How do I sort a list in Python?")
```
# Architecture
As described in more detail in the blog post, we used:
- [ALiBi](https://arxiv.org/abs/2108.12409) based attention
- [LayerNorm](https://arxiv.org/abs/1607.06450v1) instead of [RMSNorm](https://arxiv.org/pdf/1910.07467.pdf)
- [Multi Query Attention](https://arxiv.org/abs/1911.02150)
We also used LiON, flash attention, early dropout. It's not that innovative that you can't run it, in fact you can -- see an example below.
# Pretraining
For the base model, we used our own dataset that contains code with permissive licenses only, and open text datasets.
Filtering is the key to success of this model:
- We only used text in English
- Only topics related to computer science
- Applied heavy deduplication
The text to code proportion was 50:50, model trained for 1.2T tokens.
We don't release the base model, because its Fill-in-the-Middle (FIM) capability likes to repeat itself too much, so
its practical use is limited. But if you still want it, write us a message on Discord.
# Finetuning
We tested our hypothesis that chat data should boost base model performance in FIM and
regular left-to-right code completion. We found that just 15% of open
[code](https://huggingface.co/datasets/bigcode/commitpackft)
[instruction-following](https://huggingface.co/datasets/rombodawg/2XUNCENSORED_MegaCodeTraining188k) datasets,
that we filtered for quality, improves almost all metrics.
Additionally, to improve FIM, we observed common failure modes, and prepared a synthetic dataset based on
[The Stack dedup v1.1](https://huggingface.co/datasets/bigcode/the-stack-dedup) to address them.
There is a distribution shift between typical code on the internet, and the code you write in your IDE.
The former is likely finished, so the model tries to come up with a suggestion that makes the code complete.
You are likely to have half-written code as you work on it, there is no single addition that can repair it
fully.
In practice, model needs to have a tendency to stop after a couple of lines are added, and sometimes don't write
anything at all. We found that just giving it empty completions, single line completions, multiline
completions that end with a smaller text indent or at least a newline -- makes it much more usable. This data
was used as the rest 85% of the finetune dataset.
The final model is the result of several attempts to make it work as good as possible for code completion,
and to perform well on a wide range of metrics. The best attempt took 40B tokens.
# Limitations and Bias
The Refact-1.6B model was trained on text in English. But it has seen a lot more languages in
code comments. Its performance on non-English languages is lower, for sure.
# Model Stats
- **Architecture:** LLAMA-like model with multi-query attention
- **Objectives** Fill-in-the-Middle, Chat
- **Tokens context:** 4096
- **Pretraining tokens:** 1.2T
- **Finetuning tokens:** 40B
- **Precision:** bfloat16
- **GPUs** 64 NVidia A5000
- **Training time** 28 days
# License
The model is licensed under the BigScience OpenRAIL-M v1 license agreement
# Citation
If you are using this model, please give a link to this page. |
Mungert/TriLM_99M_Unpacked-GGUF | Mungert | 2025-06-15T19:40:00Z | 226 | 0 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | 2025-03-16T21:55:35Z | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_99M_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_99M_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_99M_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_99M_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_99M_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_99M_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_99M_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_99M_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_99M_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_99M_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_99M_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_99M_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# TriLM 99M Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_99M_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/Mistral-7B-Instruct-v0.1-GGUF | Mungert | 2025-06-15T19:39:56Z | 1,003 | 3 | null | [
"gguf",
"finetuned",
"text-generation",
"arxiv:2310.06825",
"base_model:mistralai/Mistral-7B-v0.1",
"base_model:quantized:mistralai/Mistral-7B-v0.1",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-16T05:47:42Z | ---
license: apache-2.0
tags:
- finetuned
base_model: mistralai/Mistral-7B-v0.1
pipeline_tag: text-generation
inference: true
widget:
- messages:
- role: user
content: What is your favorite condiment?
extra_gated_description: If you want to learn more about how we process your personal data, please read our <a href="https://mistral.ai/terms/">Privacy Policy</a>.
---
# <span style="color: #7FFF7F;">Mistral-7B-Instruct-v0.1 GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Mistral-7B-Instruct-v0.1-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Mistral-7B-Instruct-v0.1-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Mistral-7B-Instruct-v0.1-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Mistral-7B-Instruct-v0.1-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Mistral-7B-Instruct-v0.1-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Mistral-7B-Instruct-v0.1-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Mistral-7B-Instruct-v0.1-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Mistral-7B-Instruct-v0.1-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Mistral-7B-Instruct-v0.1-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Mistral-7B-Instruct-v0.1-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Mistral-7B-Instruct-v0.1-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# Model Card for Mistral-7B-Instruct-v0.1
## Encode and Decode with `mistral_common`
```py
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.protocol.instruct.messages import UserMessage
from mistral_common.protocol.instruct.request import ChatCompletionRequest
mistral_models_path = "MISTRAL_MODELS_PATH"
tokenizer = MistralTokenizer.v1()
completion_request = ChatCompletionRequest(messages=[UserMessage(content="Explain Machine Learning to me in a nutshell.")])
tokens = tokenizer.encode_chat_completion(completion_request).tokens
```
## Inference with `mistral_inference`
```py
from mistral_inference.transformer import Transformer
from mistral_inference.generate import generate
model = Transformer.from_folder(mistral_models_path)
out_tokens, _ = generate([tokens], model, max_tokens=64, temperature=0.0, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.decode(out_tokens[0])
print(result)
```
## Inference with hugging face `transformers`
```py
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.1")
model.to("cuda")
generated_ids = model.generate(tokens, max_new_tokens=1000, do_sample=True)
# decode with mistral tokenizer
result = tokenizer.decode(generated_ids[0].tolist())
print(result)
```
> [!TIP]
> PRs to correct the `transformers` tokenizer so that it gives 1-to-1 the same results as the `mistral_common` reference implementation are very welcome!
---
The Mistral-7B-Instruct-v0.1 Large Language Model (LLM) is a instruct fine-tuned version of the [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) generative text model using a variety of publicly available conversation datasets.
For full details of this model please read our [paper](https://arxiv.org/abs/2310.06825) and [release blog post](https://mistral.ai/news/announcing-mistral-7b/).
## Instruction format
In order to leverage instruction fine-tuning, your prompt should be surrounded by `[INST]` and `[/INST]` tokens. The very first instruction should begin with a begin of sentence id. The next instructions should not. The assistant generation will be ended by the end-of-sentence token id.
E.g.
```
text = "<s>[INST] What is your favourite condiment? [/INST]"
"Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s> "
"[INST] Do you have mayonnaise recipes? [/INST]"
```
This format is available as a [chat template](https://huggingface.co/docs/transformers/main/chat_templating) via the `apply_chat_template()` method:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.1")
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.1")
messages = [
{"role": "user", "content": "What is your favourite condiment?"},
{"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
{"role": "user", "content": "Do you have mayonnaise recipes?"}
]
encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt")
model_inputs = encodeds.to(device)
model.to(device)
generated_ids = model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])
```
## Model Architecture
This instruction model is based on Mistral-7B-v0.1, a transformer model with the following architecture choices:
- Grouped-Query Attention
- Sliding-Window Attention
- Byte-fallback BPE tokenizer
## Troubleshooting
- If you see the following error:
```
Traceback (most recent call last):
File "", line 1, in
File "/transformers/models/auto/auto_factory.py", line 482, in from_pretrained
config, kwargs = AutoConfig.from_pretrained(
File "/transformers/models/auto/configuration_auto.py", line 1022, in from_pretrained
config_class = CONFIG_MAPPING[config_dict["model_type"]]
File "/transformers/models/auto/configuration_auto.py", line 723, in getitem
raise KeyError(key)
KeyError: 'mistral'
```
Installing transformers from source should solve the issue
pip install git+https://github.com/huggingface/transformers
This should not be required after transformers-v4.33.4.
## Limitations
The Mistral 7B Instruct model is a quick demonstration that the base model can be easily fine-tuned to achieve compelling performance.
It does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to
make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
## The Mistral AI Team
Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed. |
Mungert/phi-2-GGUF | Mungert | 2025-06-15T19:39:50Z | 1,961 | 2 | null | [
"gguf",
"nlp",
"code",
"text-generation",
"en",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix"
] | text-generation | 2025-03-15T21:54:52Z | ---
license: mit
license_link: https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE
language:
- en
pipeline_tag: text-generation
tags:
- nlp
- code
---
# <span style="color: #7FFF7F;">phi-2 GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `phi-2-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `phi-2-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `phi-2-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `phi-2-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `phi-2-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `phi-2-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `phi-2-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `phi-2-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `phi-2-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `phi-2-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `phi-2-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
## Model Summary
Phi-2 is a Transformer with **2.7 billion** parameters. It was trained using the same data sources as [Phi-1.5](https://huggingface.co/microsoft/phi-1.5), augmented with a new data source that consists of various NLP synthetic texts and filtered websites (for safety and educational value). When assessed against benchmarks testing common sense, language understanding, and logical reasoning, Phi-2 showcased a nearly state-of-the-art performance among models with less than 13 billion parameters.
Our model hasn't been fine-tuned through reinforcement learning from human feedback. The intention behind crafting this open-source model is to provide the research community with a non-restricted small model to explore vital safety challenges, such as reducing toxicity, understanding societal biases, enhancing controllability, and more.
## How to Use
Phi-2 has been integrated in the `transformers` version 4.37.0, please ensure that you are using a version equal or higher than it.
Phi-2 is known for having an attention overflow issue (with FP16). If you are facing this issue, please enable/disable autocast on the [PhiAttention.forward()](https://github.com/huggingface/transformers/blob/main/src/transformers/models/phi/modeling_phi.py#L306) function.
## Intended Uses
Given the nature of the training data, the Phi-2 model is best suited for prompts using the QA format, the chat format, and the code format.
### QA Format:
You can provide the prompt as a standalone question as follows:
```markdown
Write a detailed analogy between mathematics and a lighthouse.
```
where the model generates the text after "." .
To encourage the model to write more concise answers, you can also try the following QA format using "Instruct: \<prompt\>\nOutput:"
```markdown
Instruct: Write a detailed analogy between mathematics and a lighthouse.
Output: Mathematics is like a lighthouse. Just as a lighthouse guides ships safely to shore, mathematics provides a guiding light in the world of numbers and logic. It helps us navigate through complex problems and find solutions. Just as a lighthouse emits a steady beam of light, mathematics provides a consistent framework for reasoning and problem-solving. It illuminates the path to understanding and helps us make sense of the world around us.
```
where the model generates the text after "Output:".
### Chat Format:
```markdown
Alice: I don't know why, I'm struggling to maintain focus while studying. Any suggestions?
Bob: Well, have you tried creating a study schedule and sticking to it?
Alice: Yes, I have, but it doesn't seem to help much.
Bob: Hmm, maybe you should try studying in a quiet environment, like the library.
Alice: ...
```
where the model generates the text after the first "Bob:".
### Code Format:
```python
def print_prime(n):
"""
Print all primes between 1 and n
"""
primes = []
for num in range(2, n+1):
is_prime = True
for i in range(2, int(math.sqrt(num))+1):
if num % i == 0:
is_prime = False
break
if is_prime:
primes.append(num)
print(primes)
```
where the model generates the text after the comments.
**Notes:**
* Phi-2 is intended for QA, chat, and code purposes. The model-generated text/code should be treated as a starting point rather than a definitive solution for potential use cases. Users should be cautious when employing these models in their applications.
* Direct adoption for production tasks without evaluation is out of scope of this project. As a result, the Phi-2 model has not been tested to ensure that it performs adequately for any production-level application. Please refer to the limitation sections of this document for more details.
* If you are using `transformers<4.37.0`, always load the model with `trust_remote_code=True` to prevent side-effects.
## Sample Code
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
torch.set_default_device("cuda")
model = AutoModelForCausalLM.from_pretrained("microsoft/phi-2", torch_dtype="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2", trust_remote_code=True)
inputs = tokenizer('''def print_prime(n):
"""
Print all primes between 1 and n
"""''', return_tensors="pt", return_attention_mask=False)
outputs = model.generate(**inputs, max_length=200)
text = tokenizer.batch_decode(outputs)[0]
print(text)
```
## Limitations of Phi-2
* Generate Inaccurate Code and Facts: The model may produce incorrect code snippets and statements. Users should treat these outputs as suggestions or starting points, not as definitive or accurate solutions.
* Limited Scope for code: Majority of Phi-2 training data is based in Python and use common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.
* Unreliable Responses to Instruction: The model has not undergone instruction fine-tuning. As a result, it may struggle or fail to adhere to intricate or nuanced instructions provided by users.
* Language Limitations: The model is primarily designed to understand standard English. Informal English, slang, or any other languages might pose challenges to its comprehension, leading to potential misinterpretations or errors in response.
* Potential Societal Biases: Phi-2 is not entirely free from societal biases despite efforts in assuring training data safety. There's a possibility it may generate content that mirrors these societal biases, particularly if prompted or instructed to do so. We urge users to be aware of this and to exercise caution and critical thinking when interpreting model outputs.
* Toxicity: Despite being trained with carefully selected data, the model can still produce harmful content if explicitly prompted or instructed to do so. We chose to release the model to help the open-source community develop the most effective ways to reduce the toxicity of a model directly after pretraining.
* Verbosity: Phi-2 being a base model often produces irrelevant or extra text and responses following its first answer to user prompts within a single turn. This is due to its training dataset being primarily textbooks, which results in textbook-like responses.
## Training
### Model
* Architecture: a Transformer-based model with next-word prediction objective
* Context length: 2048 tokens
* Dataset size: 250B tokens, combination of NLP synthetic data created by AOAI GPT-3.5 and filtered web data from Falcon RefinedWeb and SlimPajama, which was assessed by AOAI GPT-4.
* Training tokens: 1.4T tokens
* GPUs: 96xA100-80G
* Training time: 14 days
### Software
* [PyTorch](https://github.com/pytorch/pytorch)
* [DeepSpeed](https://github.com/microsoft/DeepSpeed)
* [Flash-Attention](https://github.com/HazyResearch/flash-attention)
### License
The model is licensed under the [MIT license](https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE).
## Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies. |
Mungert/Llama-3.2-3B-Instruct-GGUF | Mungert | 2025-06-15T19:39:37Z | 1,580 | 4 | transformers | [
"transformers",
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"de",
"fr",
"it",
"pt",
"hi",
"es",
"th",
"arxiv:2204.05149",
"arxiv:2405.16406",
"license:llama3.2",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-15T06:31:50Z | ---
language:
- en
- de
- fr
- it
- pt
- hi
- es
- th
library_name: transformers
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3.2
extra_gated_prompt: >-
### LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
Llama 3.2 Version Release Date: September 25, 2024
“Agreement” means the terms and conditions for use, reproduction, distribution
and modification of the Llama Materials set forth herein.
“Documentation” means the specifications, manuals and documentation accompanying Llama 3.2
distributed by Meta at https://llama.meta.com/doc/overview.
“Licensee” or “you” means you, or your employer or any other person or entity (if you are
entering into this Agreement on such person or entity’s behalf), of the age required under
applicable laws, rules or regulations to provide legal consent and that has legal authority
to bind your employer or such other person or entity if you are entering in this Agreement
on their behalf.
“Llama 3.2” means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://www.llama.com/llama-downloads.
“Llama Materials” means, collectively, Meta’s proprietary Llama 3.2 and Documentation (and
any portion thereof) made available under this Agreement.
“Meta” or “we” means Meta Platforms Ireland Limited (if you are located in or,
if you are an entity, your principal place of business is in the EEA or Switzerland)
and Meta Platforms, Inc. (if you are located outside of the EEA or Switzerland).
By clicking “I Accept” below or by using or distributing any portion or element of the Llama Materials,
you agree to be bound by this Agreement.
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide,
non-transferable and royalty-free limited license under Meta’s intellectual property or other rights
owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works
of, and make modifications to the Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works thereof),
or a product or service (including another AI model) that contains any of them, you shall (A) provide
a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Llama”
on a related website, user interface, blogpost, about page, or product documentation. If you use the
Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or
otherwise improve an AI model, which is distributed or made available, you shall also include “Llama”
at the beginning of any such AI model name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the
following attribution notice within a “Notice” text file distributed as a part of such copies:
“Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms,
Inc. All Rights Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for
the Llama Materials (available at https://www.llama.com/llama3_2/use-policy), which is hereby
incorporated by reference into this Agreement.
2. Additional Commercial Terms. If, on the Llama 3.2 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates,
is greater than 700 million monthly active users in the preceding calendar month, you must request
a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to
exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS
ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES
OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE
FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED
WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF LIABILITY,
WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT,
FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN
IF META OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama Materials,
neither Meta nor Licensee may use any name or mark owned by or associated with the other or any of its affiliates,
except as required for reasonable and customary use in describing and redistributing the Llama Materials or as
set forth in this Section 5(a). Meta hereby grants you a license to use “Llama” (the “Mark”) solely as required
to comply with the last sentence of Section 1.b.i. You will comply with Meta’s brand guidelines (currently accessible
at https://about.meta.com/brand/resources/meta/company-brand/). All goodwill arising out of your use of the Mark
will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with respect to any
derivative works and modifications of the Llama Materials that are made by you, as between you and Meta,
you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or
counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.2 outputs or results, or any portion
of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable
by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or
claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third
party arising out of or related to your use or distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this Agreement or access
to the Llama Materials and will continue in full force and effect until terminated in accordance with the terms
and conditions herein. Meta may terminate this Agreement if you are in breach of any term or condition of this
Agreement. Upon termination of this Agreement, you shall delete and cease use of the Llama Materials. Sections 3,
4 and 7 shall survive the termination of this Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of the State of
California without regard to choice of law principles, and the UN Convention on Contracts for the International
Sale of Goods does not apply to this Agreement. The courts of California shall have exclusive jurisdiction of
any dispute arising out of this Agreement.
### Llama 3.2 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.2.
If you access or use Llama 3.2, you agree to this Acceptable Use Policy (“**Policy**”).
The most recent copy of this policy can be found at
[https://www.llama.com/llama3_2/use-policy](https://www.llama.com/llama3_2/use-policy).
#### Prohibited Uses
We want everyone to use Llama 3.2 safely and responsibly. You agree you will not use, or allow others to use, Llama 3.2 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
1. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
2. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
3. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
4. Collect, process, disclose, generate, or infer private or sensitive information about individuals, including information about individuals’ identity, health, or demographic information, unless you have obtained the right to do so in accordance with applicable law
5. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
6. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
7. Engage in any action, or facilitate any action, to intentionally circumvent or remove usage restrictions or other safety measures, or to enable functionality disabled by Meta
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.2 related to the following:
8. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State or to the U.S. Biological Weapons Anti-Terrorism Act of 1989 or the Chemical Weapons Convention Implementation Act of 1997
9. Guns and illegal weapons (including weapon development)
10. Illegal drugs and regulated/controlled substances
11. Operation of critical infrastructure, transportation technologies, or heavy machinery
12. Self-harm or harm to others, including suicide, cutting, and eating disorders
13. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Llama 3.2 related to the following:
14. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
15. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
16. Generating, promoting, or further distributing spam
17. Impersonating another individual without consent, authorization, or legal right
18. Representing that the use of Llama 3.2 or outputs are human-generated
19. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
5. Interact with third party tools, models, or software designed to generate unlawful content or engage in unlawful or harmful conduct and/or represent that the outputs of such tools, models, or software are associated with Meta or Llama 3.2
With respect to any multimodal models included in Llama 3.2, the rights granted under Section 1(a) of the Llama 3.2 Community License Agreement are not being granted to you if you are an individual domiciled in, or a company with a principal place of business in, the European Union. This restriction does not apply to end users of a product or service that incorporates any such multimodal models.
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama-models/issues](https://l.workplace.com/l.php?u=https%3A%2F%2Fgithub.com%2Fmeta-llama%2Fllama-models%2Fissues&h=AT0qV8W9BFT6NwihiOHRuKYQM_UnkzN_NmHMy91OT55gkLpgi4kQupHUl0ssR4dQsIQ8n3tfd0vtkobvsEvt1l4Ic6GXI2EeuHV8N08OG2WnbAmm0FL4ObkazC6G_256vN0lN9DsykCvCqGZ)
* Reporting risky content generated by the model: [developers.facebook.com/llama_output_feedback](http://developers.facebook.com/llama_output_feedback)
* Reporting bugs and security concerns: [facebook.com/whitehat/info](http://facebook.com/whitehat/info)
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama 3.2: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Job title:
type: select
options:
- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: >-
The information you provide will be collected, stored, processed and shared in
accordance with the [Meta Privacy
Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
---
# <span style="color: #7FFF7F;">Llama-3.2-3B-Instruct GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.2-3B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.2-3B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.2-3B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.2-3B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.2-3B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.2-3B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.2-3B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.2-3B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.2-3B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.2-3B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.2-3B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
## Model Information
The Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks.
**Model Developer:** Meta
**Model Architecture:** Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
| | Training Data | Params | Input modalities | Output modalities | Context Length | GQA | Shared Embeddings | Token count | Knowledge cutoff |
| :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- |
| Llama 3.2 (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 128k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
| Llama 3.2 Quantized (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 8k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
**Supported Languages:** English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages. Developers may fine-tune Llama 3.2 models for languages beyond these supported languages, provided they comply with the Llama 3.2 Community License and the Acceptable Use Policy. Developers are always expected to ensure that their deployments, including those that involve additional languages, are completed safely and responsibly.
**Llama 3.2 Model Family:** Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date:** Sept 25, 2024
**Status:** This is a static model trained on an offline dataset. Future versions may be released that improve model capabilities and safety.
**License:** Use of Llama 3.2 is governed by the [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE) (a custom, commercial license agreement).
**Feedback:** Instructions on how to provide feedback or comments on the model can be found in the Llama Models [README](https://github.com/meta-llama/llama-models/blob/main/README.md). For more technical information about generation parameters and recipes for how to use Llama 3.2 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases:** Llama 3.2 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat and agentic applications like knowledge retrieval and summarization, mobile AI powered writing assistants and query and prompt rewriting. Pretrained models can be adapted for a variety of additional natural language generation tasks. Similarly, quantized models can be adapted for a variety of on-device use-cases with limited compute resources.
**Out of Scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3.2 Community License. Use in languages beyond those explicitly referenced as supported in this model card.
## How to use
This repository contains two versions of Llama-3.2-3B-Instruct, for use with `transformers` and with the original `llama` codebase.
### Use with transformers
Starting with `transformers >= 4.43.0` onward, you can run conversational inference using the Transformers `pipeline` abstraction or by leveraging the Auto classes with the `generate()` function.
Make sure to update your transformers installation via `pip install --upgrade transformers`.
```python
import torch
from transformers import pipeline
model_id = "meta-llama/Llama-3.2-3B-Instruct"
pipe = pipeline(
"text-generation",
model=model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
outputs = pipe(
messages,
max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])
```
Note: You can also find detailed recipes on how to use the model locally, with `torch.compile()`, assisted generations, quantised and more at [`huggingface-llama-recipes`](https://github.com/huggingface/huggingface-llama-recipes)
### Use with `llama`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama)
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Llama-3.2-3B-Instruct --include "original/*" --local-dir Llama-3.2-3B-Instruct
```
## Hardware and Software
**Training Factors:** We used custom training libraries, Meta's custom built GPU cluster, and production infrastructure for pretraining. Fine-tuning, quantization, annotation, and evaluation were also performed on production infrastructure.
**Training Energy Use:** Training utilized a cumulative of **916k** GPU hours of computation on H100-80GB (TDP of 700W) type hardware, per the table below. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
**Training Greenhouse Gas Emissions:** Estimated total location-based greenhouse gas emissions were **240** tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with renewable energy; therefore, the total market-based greenhouse gas emissions for training were 0 tons CO2eq.
| | Training Time (GPU hours) | Logit Generation Time (GPU Hours) | Training Power Consumption (W) | Training Location-Based Greenhouse Gas Emissions (tons CO2eq) | Training Market-Based Greenhouse Gas Emissions (tons CO2eq) |
| :---- | :---: | ----- | :---: | :---: | :---: |
| Llama 3.2 1B | 370k | \- | 700 | 107 | 0 |
| Llama 3.2 3B | 460k | \- | 700 | 133 | 0 |
| Llama 3.2 1B SpinQuant | 1.7 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 3B SpinQuant | 2.4 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 1B QLora | 1.3k | 0 | 700 | 0.381 | 0 |
| Llama 3.2 3B QLora | 1.6k | 0 | 700 | 0.461 | 0 |
| Total | 833k | 86k | | 240 | 0 |
\*\* The location-based CO2e emissions of Llama 3.2 1B SpinQuant and Llama 3.2 3B SpinQuant are less than 0.001 metric tonnes each. This is due to the minimal training GPU hours that are required.
The methodology used to determine training energy use and greenhouse gas emissions can be found [here](https://arxiv.org/pdf/2204.05149). Since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others.
## Training Data
**Overview:** Llama 3.2 was pretrained on up to 9 trillion tokens of data from publicly available sources. For the 1B and 3B Llama 3.2 models, we incorporated logits from the Llama 3.1 8B and 70B models into the pretraining stage of the model development, where outputs (logits) from these larger models were used as token-level targets. Knowledge distillation was used after pruning to recover performance. In post-training we used a similar recipe as Llama 3.1 and produced final chat models by doing several rounds of alignment on top of the pre-trained model. Each round involved Supervised Fine-Tuning (SFT), Rejection Sampling (RS), and Direct Preference Optimization (DPO).
**Data Freshness:** The pretraining data has a cutoff of December 2023\.
## Quantization
### Quantization Scheme
We designed the current quantization scheme with the [PyTorch’s ExecuTorch](https://github.com/pytorch/executorch) inference framework and Arm CPU backend in mind, taking into account metrics including model quality, prefill/decoding speed, and memory footprint. Our quantization scheme involves three parts:
- All linear layers in all transformer blocks are quantized to a 4-bit groupwise scheme (with a group size of 32) for weights and 8-bit per-token dynamic quantization for activations.
- The classification layer is quantized to 8-bit per-channel for weight and 8-bit per token dynamic quantization for activation.
- Similar to classification layer, an 8-bit per channel quantization is used for embedding layer.
### Quantization-Aware Training and LoRA
The quantization-aware training (QAT) with low-rank adaptation (LoRA) models went through only post-training stages, using the same data as the full precision models. To initialize QAT, we utilize BF16 Llama 3.2 model checkpoints obtained after supervised fine-tuning (SFT) and perform an additional full round of SFT training with QAT. We then freeze the backbone of the QAT model and perform another round of SFT with LoRA adaptors applied to all layers within the transformer block. Meanwhile, the LoRA adaptors' weights and activations are maintained in BF16. Because our approach is similar to QLoRA of Dettmers et al., (2023) (i.e., quantization followed by LoRA adapters), we refer this method as QLoRA. Finally, we fine-tune the resulting model (both backbone and LoRA adaptors) using direct preference optimization (DPO).
### SpinQuant
[SpinQuant](https://arxiv.org/abs/2405.16406) was applied, together with generative post-training quantization (GPTQ). For the SpinQuant rotation matrix fine-tuning, we optimized for 100 iterations, using 800 samples with sequence-length 2048 from the WikiText 2 dataset. For GPTQ, we used 128 samples from the same dataset with the same sequence-length.
## Benchmarks \- English Text
In this section, we report the results for Llama 3.2 models on standard automatic benchmarks. For all these evaluations, we used our internal evaluations library.
### Base Pretrained Models
| Category | Benchmark | \# Shots | Metric | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
| ----- | ----- | :---: | :---: | :---: | :---: | :---: |
| General | MMLU | 5 | macro\_avg/acc\_char | 32.2 | 58 | 66.7 |
| | AGIEval English | 3-5 | average/acc\_char | 23.3 | 39.2 | 47.8 |
| | ARC-Challenge | 25 | acc\_char | 32.8 | 69.1 | 79.7 |
| Reading comprehension | SQuAD | 1 | em | 49.2 | 67.7 | 77 |
| | QuAC (F1) | 1 | f1 | 37.9 | 42.9 | 44.9 |
| | DROP (F1) | 3 | f1 | 28.0 | 45.2 | 59.5 |
| Long Context | Needle in Haystack | 0 | em | 96.8 | 1 | 1 |
### Instruction Tuned Models
| Capability | | Benchmark | \# Shots | Metric | Llama 3.2 1B bf16 | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B bf16 | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | ----- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | | MMLU | 5 | macro\_avg/acc | 49.3 | 43.3 | 47.3 | 49.0 | 63.4 | 60.5 | 62 | 62.4 | 69.4 |
| Re-writing | | Open-rewrite eval | 0 | micro\_avg/rougeL | 41.6 | 39.2 | 40.9 | 41.2 | 40.1 | 40.3 | 40.8 | 40.7 | 40.9 |
| Summarization | | TLDR9+ (test) | 1 | rougeL | 16.8 | 14.9 | 16.7 | 16.8 | 19.0 | 19.1 | 19.2 | 19.1 | 17.2 |
| Instruction following | | IFEval | 0 | Avg(Prompt/Instruction acc Loose/Strict) | 59.5 | 51.5 | 58.4 | 55.6 | 77.4 | 73.9 | 73.5 | 75.9 | 80.4 |
| Math | | GSM8K (CoT) | 8 | em\_maj1@1 | 44.4 | 33.1 | 40.6 | 46.5 | 77.7 | 72.9 | 75.7 | 77.9 | 84.5 |
| | | MATH (CoT) | 0 | final\_em | 30.6 | 20.5 | 25.3 | 31.0 | 48.0 | 44.2 | 45.3 | 49.2 | 51.9 |
| Reasoning | | ARC-C | 0 | acc | 59.4 | 54.3 | 57 | 60.7 | 78.6 | 75.6 | 77.6 | 77.6 | 83.4 |
| | | GPQA | 0 | acc | 27.2 | 25.9 | 26.3 | 25.9 | 32.8 | 32.8 | 31.7 | 33.9 | 32.8 |
| | | Hellaswag | 0 | acc | 41.2 | 38.1 | 41.3 | 41.5 | 69.8 | 66.3 | 68 | 66.3 | 78.7 |
| Tool Use | | BFCL V2 | 0 | acc | 25.7 | 14.3 | 15.9 | 23.7 | 67.0 | 53.4 | 60.1 | 63.5 | 67.1 |
| | | Nexus | 0 | macro\_avg/acc | 13.5 | 5.2 | 9.6 | 12.5 | 34.3 | 32.4 | 31.5 | 30.1 | 38.5 |
| Long Context | | InfiniteBench/En.QA | 0 | longbook\_qa/f1 | 20.3 | N/A | N/A | N/A | 19.8 | N/A | N/A | N/A | 27.3 |
| | | InfiniteBench/En.MC | 0 | longbook\_choice/acc | 38.0 | N/A | N/A | N/A | 63.3 | N/A | N/A | N/A | 72.2 |
| | | NIH/Multi-needle | 0 | recall | 75.0 | N/A | N/A | N/A | 84.7 | N/A | N/A | N/A | 98.8 |
| Multilingual | | MGSM (CoT) | 0 | em | 24.5 | 13.7 | 18.2 | 24.4 | 58.2 | 48.9 | 54.3 | 56.8 | 68.9 |
\*\*for comparison purposes only. Model not released.
### Multilingual Benchmarks
| Category | Benchmark | Language | Llama 3.2 1B | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | MMLU (5-shot, macro_avg/acc) | Portuguese | 39.8 | 34.9 | 38.9 | 40.2 | 54.5 | 50.9 | 53.3 | 53.4 | 62.1 |
| | | Spanish | 41.5 | 36.0 | 39.8 | 41.8 | 55.1 | 51.9 | 53.6 | 53.6 | 62.5 |
| | | Italian | 39.8 | 34.9 | 38.1 | 40.6 | 53.8 | 49.9 | 52.1 | 51.7 | 61.6 |
| | | German | 39.2 | 34.9 | 37.5 | 39.6 | 53.3 | 50.0 | 52.2 | 51.3 | 60.6 |
| | | French | 40.5 | 34.8 | 39.2 | 40.8 | 54.6 | 51.2 | 53.3 | 53.3 | 62.3 |
| | | Hindi | 33.5 | 30.0 | 32.1 | 34.0 | 43.3 | 40.4 | 42.0 | 42.1 | 50.9 |
| | | Thai | 34.7 | 31.2 | 32.4 | 34.9 | 44.5 | 41.3 | 44.0 | 42.2 | 50.3 |
\*\*for comparison purposes only. Model not released.
## Inference time
In the below table, we compare the performance metrics of different quantization methods (SpinQuant and QAT \+ LoRA) with the BF16 baseline. The evaluation was done using the [ExecuTorch](https://github.com/pytorch/executorch) framework as the inference engine, with the ARM CPU as a backend using Android OnePlus 12 device.
| Category | Decode (tokens/sec) | Time-to-first-token (sec) | Prefill (tokens/sec) | Model size (PTE file size in MB) | Memory size (RSS in MB) |
| :---- | ----- | ----- | ----- | ----- | ----- |
| 1B BF16 (baseline) | 19.2 | 1.0 | 60.3 | 2358 | 3,185 |
| 1B SpinQuant | 50.2 (2.6x) | 0.3 (-76.9%) | 260.5 (4.3x) | 1083 (-54.1%) | 1,921 (-39.7%) |
| 1B QLoRA | 45.8 (2.4x) | 0.3 (-76.0%) | 252.0 (4.2x) | 1127 (-52.2%) | 2,255 (-29.2%) |
| 3B BF16 (baseline) | 7.6 | 3.0 | 21.2 | 6129 | 7,419 |
| 3B SpinQuant | 19.7 (2.6x) | 0.7 (-76.4%) | 89.7 (4.2x) | 2435 (-60.3%) | 3,726 (-49.8%) |
| 3B QLoRA | 18.5 (2.4x) | 0.7 (-76.1%) | 88.8 (4.2x) | 2529 (-58.7%) | 4,060 (-45.3%) |
(\*) The performance measurement is done using an adb binary-based approach.
(\*\*) It is measured on an Android OnePlus 12 device.
(\*\*\*) Time-to-first-token (TTFT) is measured with prompt length=64
*Footnote:*
- *Decode (tokens/second) is for how quickly it keeps generating. Higher is better.*
- *Time-to-first-token (TTFT for shorthand) is for how fast it generates the first token for a given prompt. Lower is better.*
- *Prefill is the inverse of TTFT (aka 1/TTFT) in tokens/second. Higher is better*
- *Model size \- how big is the model, measured by, PTE file, a binary file format for ExecuTorch*
- *RSS size \- Memory usage in resident set size (RSS)*
## Responsibility & Safety
As part of our Responsible release approach, we followed a three-pronged strategy to managing trust & safety risks:
1. Enable developers to deploy helpful, safe and flexible experiences for their target audience and for the use cases supported by Llama
2. Protect developers against adversarial users aiming to exploit Llama capabilities to potentially cause harm
3. Provide protections for the community to help prevent the misuse of our models
### Responsible Deployment
**Approach:** Llama is a foundational technology designed to be used in a variety of use cases. Examples on how Meta’s Llama models have been responsibly deployed can be found in our [Community Stories webpage](https://llama.meta.com/community-stories/). Our approach is to build the most helpful models, enabling the world to benefit from the technology power, by aligning our model safety for generic use cases and addressing a standard set of harms. Developers are then in the driver’s seat to tailor safety for their use cases, defining their own policies and deploying the models with the necessary safeguards in their Llama systems. Llama 3.2 was developed following the best practices outlined in our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/).
#### Llama 3.2 Instruct
**Objective:** Our main objectives for conducting safety fine-tuning are to provide the research community with a valuable resource for studying the robustness of safety fine-tuning, as well as to offer developers a readily available, safe, and powerful model for various applications to reduce the developer workload to deploy safe AI systems. We implemented the same set of safety mitigations as in Llama 3, and you can learn more about these in the Llama 3 [paper](https://ai.meta.com/research/publications/the-llama-3-herd-of-models/).
**Fine-Tuning Data:** We employ a multi-faceted approach to data collection, combining human-generated data from our vendors with synthetic data to mitigate potential safety risks. We’ve developed many large language model (LLM)-based classifiers that enable us to thoughtfully select high-quality prompts and responses, enhancing data quality control.
**Refusals and Tone:** Building on the work we started with Llama 3, we put a great emphasis on model refusals to benign prompts as well as refusal tone. We included both borderline and adversarial prompts in our safety data strategy, and modified our safety data responses to follow tone guidelines.
#### Llama 3.2 Systems
**Safety as a System:** Large language models, including Llama 3.2, **are not designed to be deployed in isolation** but instead should be deployed as part of an overall AI system with additional safety guardrails as required. Developers are expected to deploy system safeguards when building agentic systems. Safeguards are key to achieve the right helpfulness-safety alignment as well as mitigating safety and security risks inherent to the system and any integration of the model or system with external tools. As part of our responsible release approach, we provide the community with [safeguards](https://llama.meta.com/trust-and-safety/) that developers should deploy with Llama models or other LLMs, including Llama Guard, Prompt Guard and Code Shield. All our [reference implementations](https://github.com/meta-llama/llama-agentic-system) demos contain these safeguards by default so developers can benefit from system-level safety out-of-the-box.
### New Capabilities and Use Cases
**Technological Advancement:** Llama releases usually introduce new capabilities that require specific considerations in addition to the best practices that generally apply across all Generative AI use cases. For prior release capabilities also supported by Llama 3.2, see [Llama 3.1 Model Card](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md), as the same considerations apply here as well.
**Constrained Environments:** Llama 3.2 1B and 3B models are expected to be deployed in highly constrained environments, such as mobile devices. LLM Systems using smaller models will have a different alignment profile and safety/helpfulness tradeoff than more complex, larger systems. Developers should ensure the safety of their system meets the requirements of their use case. We recommend using lighter system safeguards for such use cases, like Llama Guard 3-1B or its mobile-optimized version.
### Evaluations
**Scaled Evaluations:** We built dedicated, adversarial evaluation datasets and evaluated systems composed of Llama models and Purple Llama safeguards to filter input prompt and output response. It is important to evaluate applications in context, and we recommend building dedicated evaluation dataset for your use case.
**Red Teaming:** We conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we used the learnings to improve our benchmarks and safety tuning datasets. We partnered early with subject-matter experts in critical risk areas to understand the nature of these real-world harms and how such models may lead to unintended harm for society. Based on these conversations, we derived a set of adversarial goals for the red team to attempt to achieve, such as extracting harmful information or reprogramming the model to act in a potentially harmful capacity. The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity in addition to multilingual content specialists with background in integrity issues in specific geographic markets.
### Critical Risks
In addition to our safety work above, we took extra care on measuring and/or mitigating the following critical risk areas:
**1\. CBRNE (Chemical, Biological, Radiological, Nuclear, and Explosive Weapons):** Llama 3.2 1B and 3B models are smaller and less capable derivatives of Llama 3.1. For Llama 3.1 70B and 405B, to assess risks related to proliferation of chemical and biological weapons, we performed uplift testing designed to assess whether use of Llama 3.1 models could meaningfully increase the capabilities of malicious actors to plan or carry out attacks using these types of weapons and have determined that such testing also applies to the smaller 1B and 3B models.
**2\. Child Safety:** Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors including the additional languages Llama 3 is trained on. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
**3\. Cyber Attacks:** For Llama 3.1 405B, our cyber attack uplift study investigated whether LLMs can enhance human capabilities in hacking tasks, both in terms of skill level and speed.
Our attack automation study focused on evaluating the capabilities of LLMs when used as autonomous agents in cyber offensive operations, specifically in the context of ransomware attacks. This evaluation was distinct from previous studies that considered LLMs as interactive assistants. The primary objective was to assess whether these models could effectively function as independent agents in executing complex cyber-attacks without human intervention. Because Llama 3.2’s 1B and 3B models are smaller and less capable models than Llama 3.1 405B, we broadly believe that the testing conducted for the 405B model also applies to Llama 3.2 models.
### Community
**Industry Partnerships:** Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership on AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
**Grants:** We also set up the [Llama Impact Grants](https://llama.meta.com/llama-impact-grants/) program to identify and support the most compelling applications of Meta’s Llama model for societal benefit across three categories: education, climate and open innovation. The 20 finalists from the hundreds of applications can be found [here](https://llama.meta.com/llama-impact-grants/#finalists).
**Reporting:** Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
**Values:** The core values of Llama 3.2 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3.2 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
**Testing:** Llama 3.2 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3.2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3.2 models, developers should perform safety testing and tuning tailored to their specific applications of the model. Please refer to available resources including our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide), [Trust and Safety](https://llama.meta.com/trust-and-safety/) solutions, and other [resources](https://llama.meta.com/docs/get-started/) to learn more about responsible development.
|
abi2736/Minds-Tutor | abi2736 | 2025-06-15T19:39:36Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T15:01:00Z | # Minds-Tutor
Projeto inicial para IA educacional. |
Mungert/Llama-3.2-1B-Instruct-GGUF | Mungert | 2025-06-15T19:39:29Z | 1,482 | 3 | transformers | [
"transformers",
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"de",
"fr",
"it",
"pt",
"hi",
"es",
"th",
"arxiv:2204.05149",
"arxiv:2405.16406",
"license:llama3.2",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-15T03:34:52Z | ---
language:
- en
- de
- fr
- it
- pt
- hi
- es
- th
library_name: transformers
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3.2
extra_gated_prompt: >-
### LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
Llama 3.2 Version Release Date: September 25, 2024
“Agreement” means the terms and conditions for use, reproduction, distribution
and modification of the Llama Materials set forth herein.
“Documentation” means the specifications, manuals and documentation accompanying Llama 3.2
distributed by Meta at https://llama.meta.com/doc/overview.
“Licensee” or “you” means you, or your employer or any other person or entity (if you are
entering into this Agreement on such person or entity’s behalf), of the age required under
applicable laws, rules or regulations to provide legal consent and that has legal authority
to bind your employer or such other person or entity if you are entering in this Agreement
on their behalf.
“Llama 3.2” means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://www.llama.com/llama-downloads.
“Llama Materials” means, collectively, Meta’s proprietary Llama 3.2 and Documentation (and
any portion thereof) made available under this Agreement.
“Meta” or “we” means Meta Platforms Ireland Limited (if you are located in or,
if you are an entity, your principal place of business is in the EEA or Switzerland)
and Meta Platforms, Inc. (if you are located outside of the EEA or Switzerland).
By clicking “I Accept” below or by using or distributing any portion or element of the Llama Materials,
you agree to be bound by this Agreement.
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide,
non-transferable and royalty-free limited license under Meta’s intellectual property or other rights
owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works
of, and make modifications to the Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works thereof),
or a product or service (including another AI model) that contains any of them, you shall (A) provide
a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Llama”
on a related website, user interface, blogpost, about page, or product documentation. If you use the
Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or
otherwise improve an AI model, which is distributed or made available, you shall also include “Llama”
at the beginning of any such AI model name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the
following attribution notice within a “Notice” text file distributed as a part of such copies:
“Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms,
Inc. All Rights Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for
the Llama Materials (available at https://www.llama.com/llama3_2/use-policy), which is hereby
incorporated by reference into this Agreement.
2. Additional Commercial Terms. If, on the Llama 3.2 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates,
is greater than 700 million monthly active users in the preceding calendar month, you must request
a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to
exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS
ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES
OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE
FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED
WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF LIABILITY,
WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT,
FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN
IF META OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama Materials,
neither Meta nor Licensee may use any name or mark owned by or associated with the other or any of its affiliates,
except as required for reasonable and customary use in describing and redistributing the Llama Materials or as
set forth in this Section 5(a). Meta hereby grants you a license to use “Llama” (the “Mark”) solely as required
to comply with the last sentence of Section 1.b.i. You will comply with Meta’s brand guidelines (currently accessible
at https://about.meta.com/brand/resources/meta/company-brand/). All goodwill arising out of your use of the Mark
will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with respect to any
derivative works and modifications of the Llama Materials that are made by you, as between you and Meta,
you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or
counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.2 outputs or results, or any portion
of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable
by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or
claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third
party arising out of or related to your use or distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this Agreement or access
to the Llama Materials and will continue in full force and effect until terminated in accordance with the terms
and conditions herein. Meta may terminate this Agreement if you are in breach of any term or condition of this
Agreement. Upon termination of this Agreement, you shall delete and cease use of the Llama Materials. Sections 3,
4 and 7 shall survive the termination of this Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of the State of
California without regard to choice of law principles, and the UN Convention on Contracts for the International
Sale of Goods does not apply to this Agreement. The courts of California shall have exclusive jurisdiction of
any dispute arising out of this Agreement.
### Llama 3.2 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.2.
If you access or use Llama 3.2, you agree to this Acceptable Use Policy (“**Policy**”).
The most recent copy of this policy can be found at
[https://www.llama.com/llama3_2/use-policy](https://www.llama.com/llama3_2/use-policy).
#### Prohibited Uses
We want everyone to use Llama 3.2 safely and responsibly. You agree you will not use, or allow others to use, Llama 3.2 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
1. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
2. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
3. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
4. Collect, process, disclose, generate, or infer private or sensitive information about individuals, including information about individuals’ identity, health, or demographic information, unless you have obtained the right to do so in accordance with applicable law
5. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
6. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
7. Engage in any action, or facilitate any action, to intentionally circumvent or remove usage restrictions or other safety measures, or to enable functionality disabled by Meta
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.2 related to the following:
8. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State or to the U.S. Biological Weapons Anti-Terrorism Act of 1989 or the Chemical Weapons Convention Implementation Act of 1997
9. Guns and illegal weapons (including weapon development)
10. Illegal drugs and regulated/controlled substances
11. Operation of critical infrastructure, transportation technologies, or heavy machinery
12. Self-harm or harm to others, including suicide, cutting, and eating disorders
13. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Llama 3.2 related to the following:
14. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
15. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
16. Generating, promoting, or further distributing spam
17. Impersonating another individual without consent, authorization, or legal right
18. Representing that the use of Llama 3.2 or outputs are human-generated
19. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
5. Interact with third party tools, models, or software designed to generate unlawful content or engage in unlawful or harmful conduct and/or represent that the outputs of such tools, models, or software are associated with Meta or Llama 3.2
With respect to any multimodal models included in Llama 3.2, the rights granted under Section 1(a) of the Llama 3.2 Community License Agreement are not being granted to you if you are an individual domiciled in, or a company with a principal place of business in, the European Union. This restriction does not apply to end users of a product or service that incorporates any such multimodal models.
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama-models/issues](https://l.workplace.com/l.php?u=https%3A%2F%2Fgithub.com%2Fmeta-llama%2Fllama-models%2Fissues&h=AT0qV8W9BFT6NwihiOHRuKYQM_UnkzN_NmHMy91OT55gkLpgi4kQupHUl0ssR4dQsIQ8n3tfd0vtkobvsEvt1l4Ic6GXI2EeuHV8N08OG2WnbAmm0FL4ObkazC6G_256vN0lN9DsykCvCqGZ)
* Reporting risky content generated by the model: [developers.facebook.com/llama_output_feedback](http://developers.facebook.com/llama_output_feedback)
* Reporting bugs and security concerns: [facebook.com/whitehat/info](http://facebook.com/whitehat/info)
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama 3.2: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Job title:
type: select
options:
- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: >-
The information you provide will be collected, stored, processed and shared in
accordance with the [Meta Privacy
Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
---
# <span style="color: #7FFF7F;">Llama-3.2-1B-Instruct GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.2-1B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.2-1B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.2-1B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.2-1B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.2-1B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.2-1B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.2-1B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.2-1B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.2-1B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.2-1B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.2-1B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
## Model Information
The Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks.
**Model Developer:** Meta
**Model Architecture:** Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
| | Training Data | Params | Input modalities | Output modalities | Context Length | GQA | Shared Embeddings | Token count | Knowledge cutoff |
| :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- |
| Llama 3.2 (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 128k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
| Llama 3.2 Quantized (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 8k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
**Supported Languages:** English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages. Developers may fine-tune Llama 3.2 models for languages beyond these supported languages, provided they comply with the Llama 3.2 Community License and the Acceptable Use Policy. Developers are always expected to ensure that their deployments, including those that involve additional languages, are completed safely and responsibly.
**Llama 3.2 Model Family:** Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date:** Sept 25, 2024
**Status:** This is a static model trained on an offline dataset. Future versions may be released that improve model capabilities and safety.
**License:** Use of Llama 3.2 is governed by the [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE) (a custom, commercial license agreement).
**Feedback:** Instructions on how to provide feedback or comments on the model can be found in the Llama Models [README](https://github.com/meta-llama/llama-models/blob/main/README.md). For more technical information about generation parameters and recipes for how to use Llama 3.2 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases:** Llama 3.2 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat and agentic applications like knowledge retrieval and summarization, mobile AI powered writing assistants and query and prompt rewriting. Pretrained models can be adapted for a variety of additional natural language generation tasks. Similarly, quantized models can be adapted for a variety of on-device use-cases with limited compute resources.
**Out of Scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3.2 Community License. Use in languages beyond those explicitly referenced as supported in this model card.
## How to use
This repository contains two versions of Llama-3.2-1B-Instruct, for use with transformers and with the original `llama` codebase.
### Use with transformers
Starting with `transformers >= 4.43.0` onward, you can run conversational inference using the Transformers `pipeline` abstraction or by leveraging the Auto classes with the `generate()` function.
Make sure to update your transformers installation via `pip install --upgrade transformers`.
```python
import torch
from transformers import pipeline
model_id = "meta-llama/Llama-3.2-1B-Instruct"
pipe = pipeline(
"text-generation",
model=model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
outputs = pipe(
messages,
max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])
```
Note: You can also find detailed recipes on how to use the model locally, with `torch.compile()`, assisted generations, quantised and more at [`huggingface-llama-recipes`](https://github.com/huggingface/huggingface-llama-recipes)
### Use with `llama`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama)
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Llama-3.2-1B-Instruct --include "original/*" --local-dir Llama-3.2-1B-Instruct
```
## Hardware and Software
**Training Factors:** We used custom training libraries, Meta's custom built GPU cluster, and production infrastructure for pretraining. Fine-tuning, quantization, annotation, and evaluation were also performed on production infrastructure.
**Training Energy Use:** Training utilized a cumulative of **916k** GPU hours of computation on H100-80GB (TDP of 700W) type hardware, per the table below. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
**Training Greenhouse Gas Emissions:** Estimated total location-based greenhouse gas emissions were **240** tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with renewable energy; therefore, the total market-based greenhouse gas emissions for training were 0 tons CO2eq.
| | Training Time (GPU hours) | Logit Generation Time (GPU Hours) | Training Power Consumption (W) | Training Location-Based Greenhouse Gas Emissions (tons CO2eq) | Training Market-Based Greenhouse Gas Emissions (tons CO2eq) |
| :---- | :---: | ----- | :---: | :---: | :---: |
| Llama 3.2 1B | 370k | \- | 700 | 107 | 0 |
| Llama 3.2 3B | 460k | \- | 700 | 133 | 0 |
| Llama 3.2 1B SpinQuant | 1.7 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 3B SpinQuant | 2.4 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 1B QLora | 1.3k | 0 | 700 | 0.381 | 0 |
| Llama 3.2 3B QLora | 1.6k | 0 | 700 | 0.461 | 0 |
| Total | 833k | 86k | | 240 | 0 |
\*\* The location-based CO2e emissions of Llama 3.2 1B SpinQuant and Llama 3.2 3B SpinQuant are less than 0.001 metric tonnes each. This is due to the minimal training GPU hours that are required.
The methodology used to determine training energy use and greenhouse gas emissions can be found [here](https://arxiv.org/pdf/2204.05149). Since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others.
## Training Data
**Overview:** Llama 3.2 was pretrained on up to 9 trillion tokens of data from publicly available sources. For the 1B and 3B Llama 3.2 models, we incorporated logits from the Llama 3.1 8B and 70B models into the pretraining stage of the model development, where outputs (logits) from these larger models were used as token-level targets. Knowledge distillation was used after pruning to recover performance. In post-training we used a similar recipe as Llama 3.1 and produced final chat models by doing several rounds of alignment on top of the pre-trained model. Each round involved Supervised Fine-Tuning (SFT), Rejection Sampling (RS), and Direct Preference Optimization (DPO).
**Data Freshness:** The pretraining data has a cutoff of December 2023\.
## Quantization
### Quantization Scheme
We designed the current quantization scheme with the [PyTorch’s ExecuTorch](https://github.com/pytorch/executorch) inference framework and Arm CPU backend in mind, taking into account metrics including model quality, prefill/decoding speed, and memory footprint. Our quantization scheme involves three parts:
- All linear layers in all transformer blocks are quantized to a 4-bit groupwise scheme (with a group size of 32) for weights and 8-bit per-token dynamic quantization for activations.
- The classification layer is quantized to 8-bit per-channel for weight and 8-bit per token dynamic quantization for activation.
- Similar to classification layer, an 8-bit per channel quantization is used for embedding layer.
### Quantization-Aware Training and LoRA
The quantization-aware training (QAT) with low-rank adaptation (LoRA) models went through only post-training stages, using the same data as the full precision models. To initialize QAT, we utilize BF16 Llama 3.2 model checkpoints obtained after supervised fine-tuning (SFT) and perform an additional full round of SFT training with QAT. We then freeze the backbone of the QAT model and perform another round of SFT with LoRA adaptors applied to all layers within the transformer block. Meanwhile, the LoRA adaptors' weights and activations are maintained in BF16. Because our approach is similar to QLoRA of Dettmers et al., (2023) (i.e., quantization followed by LoRA adapters), we refer this method as QLoRA. Finally, we fine-tune the resulting model (both backbone and LoRA adaptors) using direct preference optimization (DPO).
### SpinQuant
[SpinQuant](https://arxiv.org/abs/2405.16406) was applied, together with generative post-training quantization (GPTQ). For the SpinQuant rotation matrix fine-tuning, we optimized for 100 iterations, using 800 samples with sequence-length 2048 from the WikiText 2 dataset. For GPTQ, we used 128 samples from the same dataset with the same sequence-length.
## Benchmarks \- English Text
In this section, we report the results for Llama 3.2 models on standard automatic benchmarks. For all these evaluations, we used our internal evaluations library.
### Base Pretrained Models
| Category | Benchmark | \# Shots | Metric | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
| ----- | ----- | :---: | :---: | :---: | :---: | :---: |
| General | MMLU | 5 | macro\_avg/acc\_char | 32.2 | 58 | 66.7 |
| | AGIEval English | 3-5 | average/acc\_char | 23.3 | 39.2 | 47.8 |
| | ARC-Challenge | 25 | acc\_char | 32.8 | 69.1 | 79.7 |
| Reading comprehension | SQuAD | 1 | em | 49.2 | 67.7 | 77 |
| | QuAC (F1) | 1 | f1 | 37.9 | 42.9 | 44.9 |
| | DROP (F1) | 3 | f1 | 28.0 | 45.2 | 59.5 |
| Long Context | Needle in Haystack | 0 | em | 96.8 | 1 | 1 |
### Instruction Tuned Models
| Capability | | Benchmark | \# Shots | Metric | Llama 3.2 1B bf16 | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B bf16 | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | ----- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | | MMLU | 5 | macro\_avg/acc | 49.3 | 43.3 | 47.3 | 49.0 | 63.4 | 60.5 | 62 | 62.4 | 69.4 |
| Re-writing | | Open-rewrite eval | 0 | micro\_avg/rougeL | 41.6 | 39.2 | 40.9 | 41.2 | 40.1 | 40.3 | 40.8 | 40.7 | 40.9 |
| Summarization | | TLDR9+ (test) | 1 | rougeL | 16.8 | 14.9 | 16.7 | 16.8 | 19.0 | 19.1 | 19.2 | 19.1 | 17.2 |
| Instruction following | | IFEval | 0 | Avg(Prompt/Instruction acc Loose/Strict) | 59.5 | 51.5 | 58.4 | 55.6 | 77.4 | 73.9 | 73.5 | 75.9 | 80.4 |
| Math | | GSM8K (CoT) | 8 | em\_maj1@1 | 44.4 | 33.1 | 40.6 | 46.5 | 77.7 | 72.9 | 75.7 | 77.9 | 84.5 |
| | | MATH (CoT) | 0 | final\_em | 30.6 | 20.5 | 25.3 | 31.0 | 48.0 | 44.2 | 45.3 | 49.2 | 51.9 |
| Reasoning | | ARC-C | 0 | acc | 59.4 | 54.3 | 57 | 60.7 | 78.6 | 75.6 | 77.6 | 77.6 | 83.4 |
| | | GPQA | 0 | acc | 27.2 | 25.9 | 26.3 | 25.9 | 32.8 | 32.8 | 31.7 | 33.9 | 32.8 |
| | | Hellaswag | 0 | acc | 41.2 | 38.1 | 41.3 | 41.5 | 69.8 | 66.3 | 68 | 66.3 | 78.7 |
| Tool Use | | BFCL V2 | 0 | acc | 25.7 | 14.3 | 15.9 | 23.7 | 67.0 | 53.4 | 60.1 | 63.5 | 67.1 |
| | | Nexus | 0 | macro\_avg/acc | 13.5 | 5.2 | 9.6 | 12.5 | 34.3 | 32.4 | 31.5 | 30.1 | 38.5 |
| Long Context | | InfiniteBench/En.QA | 0 | longbook\_qa/f1 | 20.3 | N/A | N/A | N/A | 19.8 | N/A | N/A | N/A | 27.3 |
| | | InfiniteBench/En.MC | 0 | longbook\_choice/acc | 38.0 | N/A | N/A | N/A | 63.3 | N/A | N/A | N/A | 72.2 |
| | | NIH/Multi-needle | 0 | recall | 75.0 | N/A | N/A | N/A | 84.7 | N/A | N/A | N/A | 98.8 |
| Multilingual | | MGSM (CoT) | 0 | em | 24.5 | 13.7 | 18.2 | 24.4 | 58.2 | 48.9 | 54.3 | 56.8 | 68.9 |
\*\*for comparison purposes only. Model not released.
### Multilingual Benchmarks
| Category | Benchmark | Language | Llama 3.2 1B | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | MMLU (5-shot, macro_avg/acc) | Portuguese | 39.8 | 34.9 | 38.9 | 40.2 | 54.5 | 50.9 | 53.3 | 53.4 | 62.1 |
| | | Spanish | 41.5 | 36.0 | 39.8 | 41.8 | 55.1 | 51.9 | 53.6 | 53.6 | 62.5 |
| | | Italian | 39.8 | 34.9 | 38.1 | 40.6 | 53.8 | 49.9 | 52.1 | 51.7 | 61.6 |
| | | German | 39.2 | 34.9 | 37.5 | 39.6 | 53.3 | 50.0 | 52.2 | 51.3 | 60.6 |
| | | French | 40.5 | 34.8 | 39.2 | 40.8 | 54.6 | 51.2 | 53.3 | 53.3 | 62.3 |
| | | Hindi | 33.5 | 30.0 | 32.1 | 34.0 | 43.3 | 40.4 | 42.0 | 42.1 | 50.9 |
| | | Thai | 34.7 | 31.2 | 32.4 | 34.9 | 44.5 | 41.3 | 44.0 | 42.2 | 50.3 |
\*\*for comparison purposes only. Model not released.
## Inference time
In the below table, we compare the performance metrics of different quantization methods (SpinQuant and QAT \+ LoRA) with the BF16 baseline. The evaluation was done using the [ExecuTorch](https://github.com/pytorch/executorch) framework as the inference engine, with the ARM CPU as a backend using Android OnePlus 12 device.
| Category | Decode (tokens/sec) | Time-to-first-token (sec) | Prefill (tokens/sec) | Model size (PTE file size in MB) | Memory size (RSS in MB) |
| :---- | ----- | ----- | ----- | ----- | ----- |
| 1B BF16 (baseline) | 19.2 | 1.0 | 60.3 | 2358 | 3,185 |
| 1B SpinQuant | 50.2 (2.6x) | 0.3 (-76.9%) | 260.5 (4.3x) | 1083 (-54.1%) | 1,921 (-39.7%) |
| 1B QLoRA | 45.8 (2.4x) | 0.3 (-76.0%) | 252.0 (4.2x) | 1127 (-52.2%) | 2,255 (-29.2%) |
| 3B BF16 (baseline) | 7.6 | 3.0 | 21.2 | 6129 | 7,419 |
| 3B SpinQuant | 19.7 (2.6x) | 0.7 (-76.4%) | 89.7 (4.2x) | 2435 (-60.3%) | 3,726 (-49.8%) |
| 3B QLoRA | 18.5 (2.4x) | 0.7 (-76.1%) | 88.8 (4.2x) | 2529 (-58.7%) | 4,060 (-45.3%) |
(\*) The performance measurement is done using an adb binary-based approach.
(\*\*) It is measured on an Android OnePlus 12 device.
(\*\*\*) Time-to-first-token (TTFT) is measured with prompt length=64
*Footnote:*
- *Decode (tokens/second) is for how quickly it keeps generating. Higher is better.*
- *Time-to-first-token (TTFT for shorthand) is for how fast it generates the first token for a given prompt. Lower is better.*
- *Prefill is the inverse of TTFT (aka 1/TTFT) in tokens/second. Higher is better*
- *Model size \- how big is the model, measured by, PTE file, a binary file format for ExecuTorch*
- *RSS size \- Memory usage in resident set size (RSS)*
## Responsibility & Safety
As part of our Responsible release approach, we followed a three-pronged strategy to managing trust & safety risks:
1. Enable developers to deploy helpful, safe and flexible experiences for their target audience and for the use cases supported by Llama
2. Protect developers against adversarial users aiming to exploit Llama capabilities to potentially cause harm
3. Provide protections for the community to help prevent the misuse of our models
### Responsible Deployment
**Approach:** Llama is a foundational technology designed to be used in a variety of use cases. Examples on how Meta’s Llama models have been responsibly deployed can be found in our [Community Stories webpage](https://llama.meta.com/community-stories/). Our approach is to build the most helpful models, enabling the world to benefit from the technology power, by aligning our model safety for generic use cases and addressing a standard set of harms. Developers are then in the driver’s seat to tailor safety for their use cases, defining their own policies and deploying the models with the necessary safeguards in their Llama systems. Llama 3.2 was developed following the best practices outlined in our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/).
#### Llama 3.2 Instruct
**Objective:** Our main objectives for conducting safety fine-tuning are to provide the research community with a valuable resource for studying the robustness of safety fine-tuning, as well as to offer developers a readily available, safe, and powerful model for various applications to reduce the developer workload to deploy safe AI systems. We implemented the same set of safety mitigations as in Llama 3, and you can learn more about these in the Llama 3 [paper](https://ai.meta.com/research/publications/the-llama-3-herd-of-models/).
**Fine-Tuning Data:** We employ a multi-faceted approach to data collection, combining human-generated data from our vendors with synthetic data to mitigate potential safety risks. We’ve developed many large language model (LLM)-based classifiers that enable us to thoughtfully select high-quality prompts and responses, enhancing data quality control.
**Refusals and Tone:** Building on the work we started with Llama 3, we put a great emphasis on model refusals to benign prompts as well as refusal tone. We included both borderline and adversarial prompts in our safety data strategy, and modified our safety data responses to follow tone guidelines.
#### Llama 3.2 Systems
**Safety as a System:** Large language models, including Llama 3.2, **are not designed to be deployed in isolation** but instead should be deployed as part of an overall AI system with additional safety guardrails as required. Developers are expected to deploy system safeguards when building agentic systems. Safeguards are key to achieve the right helpfulness-safety alignment as well as mitigating safety and security risks inherent to the system and any integration of the model or system with external tools. As part of our responsible release approach, we provide the community with [safeguards](https://llama.meta.com/trust-and-safety/) that developers should deploy with Llama models or other LLMs, including Llama Guard, Prompt Guard and Code Shield. All our [reference implementations](https://github.com/meta-llama/llama-agentic-system) demos contain these safeguards by default so developers can benefit from system-level safety out-of-the-box.
### New Capabilities and Use Cases
**Technological Advancement:** Llama releases usually introduce new capabilities that require specific considerations in addition to the best practices that generally apply across all Generative AI use cases. For prior release capabilities also supported by Llama 3.2, see [Llama 3.1 Model Card](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md), as the same considerations apply here as well.
**Constrained Environments:** Llama 3.2 1B and 3B models are expected to be deployed in highly constrained environments, such as mobile devices. LLM Systems using smaller models will have a different alignment profile and safety/helpfulness tradeoff than more complex, larger systems. Developers should ensure the safety of their system meets the requirements of their use case. We recommend using lighter system safeguards for such use cases, like Llama Guard 3-1B or its mobile-optimized version.
### Evaluations
**Scaled Evaluations:** We built dedicated, adversarial evaluation datasets and evaluated systems composed of Llama models and Purple Llama safeguards to filter input prompt and output response. It is important to evaluate applications in context, and we recommend building dedicated evaluation dataset for your use case.
**Red Teaming:** We conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we used the learnings to improve our benchmarks and safety tuning datasets. We partnered early with subject-matter experts in critical risk areas to understand the nature of these real-world harms and how such models may lead to unintended harm for society. Based on these conversations, we derived a set of adversarial goals for the red team to attempt to achieve, such as extracting harmful information or reprogramming the model to act in a potentially harmful capacity. The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity in addition to multilingual content specialists with background in integrity issues in specific geographic markets.
### Critical Risks
In addition to our safety work above, we took extra care on measuring and/or mitigating the following critical risk areas:
**1\. CBRNE (Chemical, Biological, Radiological, Nuclear, and Explosive Weapons):** Llama 3.2 1B and 3B models are smaller and less capable derivatives of Llama 3.1. For Llama 3.1 70B and 405B, to assess risks related to proliferation of chemical and biological weapons, we performed uplift testing designed to assess whether use of Llama 3.1 models could meaningfully increase the capabilities of malicious actors to plan or carry out attacks using these types of weapons and have determined that such testing also applies to the smaller 1B and 3B models.
**2\. Child Safety:** Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors including the additional languages Llama 3 is trained on. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
**3\. Cyber Attacks:** For Llama 3.1 405B, our cyber attack uplift study investigated whether LLMs can enhance human capabilities in hacking tasks, both in terms of skill level and speed.
Our attack automation study focused on evaluating the capabilities of LLMs when used as autonomous agents in cyber offensive operations, specifically in the context of ransomware attacks. This evaluation was distinct from previous studies that considered LLMs as interactive assistants. The primary objective was to assess whether these models could effectively function as independent agents in executing complex cyber-attacks without human intervention. Because Llama 3.2’s 1B and 3B models are smaller and less capable models than Llama 3.1 405B, we broadly believe that the testing conducted for the 405B model also applies to Llama 3.2 models.
### Community
**Industry Partnerships:** Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership on AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
**Grants:** We also set up the [Llama Impact Grants](https://llama.meta.com/llama-impact-grants/) program to identify and support the most compelling applications of Meta’s Llama model for societal benefit across three categories: education, climate and open innovation. The 20 finalists from the hundreds of applications can be found [here](https://llama.meta.com/llama-impact-grants/#finalists).
**Reporting:** Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
**Values:** The core values of Llama 3.2 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3.2 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
**Testing:** Llama 3.2 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3.2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3.2 models, developers should perform safety testing and tuning tailored to their specific applications of the model. Please refer to available resources including our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide), [Trust and Safety](https://llama.meta.com/trust-and-safety/) solutions, and other [resources](https://llama.meta.com/docs/get-started/) to learn more about responsible development.
|
Mungert/gemma-3-12b-it-gguf | Mungert | 2025-06-15T19:39:16Z | 3,675 | 11 | null | [
"gguf",
"gemma",
"vision",
"image",
"llama.cpp",
"image-text-to-text",
"license:gemma",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | image-text-to-text | 2025-03-12T22:44:45Z | ---
license: gemma
pipeline_tag: image-text-to-text
tags:
- gemma
- vision
- image
- llama.cpp
---
# <span style="color: #7FFF7F;">Gemma-3 12B Instruct GGUF Models</span>
## How to Use Gemma 3 Vision with llama.cpp
To utilize the experimental support for Gemma 3 Vision in `llama.cpp`, follow these steps:
1. **Clone the lastest llama.cpp Repository**:
```bash
git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
```
2. **Build the Llama.cpp**:
Build llama.cpp as usual : https://github.com/ggml-org/llama.cpp#building-the-project
Once llama.cpp is built Copy the ./llama.cpp/build/bin/llama-gemma3-cli to a chosen folder.
3. **Download the Gemma 3 gguf file**:
https://huggingface.co/Mungert/gemma-3-12b-it-gguf/tree/main
Choose a gguf file without the mmproj in the name
Example gguf file : https://huggingface.co/Mungert/gemma-3-12b-it-gguf/resolve/main/google_gemma-3-12b-it-q4_k_l.gguf
Copy this file to your chosen folder.
4. **Download the Gemma 3 mmproj file**
https://huggingface.co/Mungert/gemma-3-12b-it-gguf/tree/main
Choose a file with mmproj in the name
Example mmproj file : https://huggingface.co/Mungert/gemma-3-12b-it-gguf/resolve/main/google_gemma-3-12b-it-mmproj-bf16.gguf
Copy this file to your chosen folder.
5. Copy images to the same folder as the gguf files or alter paths appropriately.
In the example below the gguf files, images and llama-gemma-cli are in the same folder.
Example image: image https://huggingface.co/Mungert/gemma-3-12b-it-gguf/resolve/main/car-1.jpg
Copy this file to your chosen folder.
6. **Run the CLI Tool**:
From your chosen folder :
```bash
llama-gemma3-cli -m google_gemma-3-12b-it-q4_k_l.gguf --mmproj google_gemma-3-12b-it-mmproj-bf16.gguf
```
```
Running in chat mode, available commands:
/image <path> load an image
/clear clear the chat history
/quit or /exit exit the program
> /image car-1.jpg
Encoding image car-1.jpg
Image encoded in 46305 ms
Image decoded in 19302 ms
> what is the image of
Here's a breakdown of what's in the image:
**Subject:** The primary subject is a black Porsche Panamera Turbo driving on a highway.
**Details:**
* **Car:** It's a sleek, modern Porsche Panamera Turbo, identifiable by its distinctive rear design, the "PORSCHE" lettering, and the "Panamera Turbo" badge. The license plate reads "CVC-911".
* **Setting:** The car is on a multi-lane highway, with a blurred background of trees, a distant building, and a cloudy sky. The lighting suggests it's either dusk or dawn.
* **Motion:** The image captures the car in motion, with a slight motion blur to convey speed.
**Overall Impression:** The image conveys a sense of speed, luxury, and power. It's a well-composed shot that highlights the car's design and performance.
Do you want me to describe any specific aspect of the image in more detail, or perhaps analyze its composition?
```
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤️ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs **Phi-4-mini-instruct** using phi-4-mini-q4_0.gguf , llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limtations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Low | Very Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium Low | Low | CPU with more memory | Better accuracy while still being quantized |
| **Q8** | Medium | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
## **Included Files & Details**
### `google_gemma-3-12b-it-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `google_gemma-3-12b-it-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `google_gemma-3-12b-it-bf16-q8.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `google_gemma-3-12b-it-f16-q8.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `google_gemma-3-12b-it-q4_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `google_gemma-3-12b-it-q4_k_m.gguf`
- Similar to Q4_K.
- Another option for **low-memory CPU inference**.
### `google_gemma-3-12b-it-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `google_gemma-3-12b-it-q6_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `google_gemma-3-12b-it-q6_k_m.gguf`
- A mid-range **Q6_K** quantized model for balanced performance .
- Suitable for **CPU-based inference** with **moderate memory**.
### `google_gemma-3-12b-it-q8.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
# Gemma 3 model card
**Model Page**: [Gemma](https://ai.google.dev/gemma/docs/core)
**Resources and Technical Documentation**:
* [Gemma 3 Technical Report][g3-tech-report]
* [Responsible Generative AI Toolkit][rai-toolkit]
* [Gemma on Kaggle][kaggle-gemma]
* [Gemma on Vertex Model Garden][vertex-mg-gemma3]
**Terms of Use**: [Terms][terms]
**Authors**: Google DeepMind
## Model Information
Summary description and brief definition of inputs and outputs.
### Description
Gemma is a family of lightweight, state-of-the-art open models from Google,
built from the same research and technology used to create the Gemini models.
Gemma 3 models are multimodal, handling text and image input and generating text
output, with open weights for both pre-trained variants and instruction-tuned
variants. Gemma 3 has a large, 128K context window, multilingual support in over
140 languages, and is available in more sizes than previous versions. Gemma 3
models are well-suited for a variety of text generation and image understanding
tasks, including question answering, summarization, and reasoning. Their
relatively small size makes it possible to deploy them in environments with
limited resources such as laptops, desktops or your own cloud infrastructure,
democratizing access to state of the art AI models and helping foster innovation
for everyone.
### Inputs and outputs
- **Input:**
- Text string, such as a question, a prompt, or a document to be summarized
- Images, normalized to 896 x 896 resolution and encoded to 256 tokens
each
- Total input context of 128K tokens for the 4B, 12B, and 27B sizes, and
32K tokens for the 1B size
- **Output:**
- Generated text in response to the input, such as an answer to a
question, analysis of image content, or a summary of a document
- Total output context of 8192 tokens
|
Mungert/gemma-3-4b-it-gguf | Mungert | 2025-06-15T19:39:13Z | 1,145 | 12 | null | [
"gguf",
"vision",
"gemma",
"llama.cpp",
"image-text-to-text",
"license:gemma",
"endpoints_compatible",
"region:us",
"conversational"
] | image-text-to-text | 2025-03-12T19:18:29Z | ---
license: gemma
pipeline_tag: image-text-to-text
tags:
- vision
- gemma
- llama.cpp
---
# <span style="color: #7FFF7F;">Gemma-3 4B Instruct GGUF Models</span>
## How to Use Gemma 3 Vision with llama.cpp
To utilize the experimental support for Gemma 3 Vision in `llama.cpp`, follow these steps:
1. **Clone the lastest llama.cpp Repository**:
```bash
git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
```
2. **Build the Llama.cpp**:
Build llama.cpp as usual : https://github.com/ggml-org/llama.cpp#building-the-project
Once llama.cpp is built Copy the ./llama.cpp/build/bin/llama-gemma3-cli to a chosen folder.
3. **Download the Gemma 3 gguf file**:
https://huggingface.co/Mungert/gemma-3-4b-it-gguf/tree/main
Choose a gguf file without the mmproj in the name
Example gguf file : https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/google_gemma-3-4b-it-q4_k_l.gguf
Copy this file to your chosen folder.
4. **Download the Gemma 3 mmproj file**
https://huggingface.co/Mungert/gemma-3-4b-it-gguf/tree/main
Choose a file with mmproj in the name
Example mmproj file : https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/google_gemma-3-4b-it-mmproj-bf16.gguf
Copy this file to your chosen folder.
5. Copy images to the same folder as the gguf files or alter paths appropriately.
In the example below the gguf files, images and llama-gemma-cli are in the same folder.
Example image: image https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/car-1.jpg
Copy this file to your chosen folder.
6. **Run the CLI Tool**:
From your chosen folder :
```bash
llama-gemma3-cli -m google_gemma-3-4b-it-q4_k_l.gguf --mmproj google_gemma-3-4b-it-mmproj-bf16.gguf
```
```
Running in chat mode, available commands:
/image <path> load an image
/clear clear the chat history
/quit or /exit exit the program
> /image car-1.jpg
Encoding image car-1.jpg
Image encoded in 46305 ms
Image decoded in 19302 ms
> what is the image of
Here's a breakdown of what's in the image:
**Subject:** The primary subject is a black Porsche Panamera Turbo driving on a highway.
**Details:**
* **Car:** It's a sleek, modern Porsche Panamera Turbo, identifiable by its distinctive rear design, the "PORSCHE" lettering, and the "Panamera Turbo" badge. The license plate reads "CVC-911".
* **Setting:** The car is on a multi-lane highway, with a blurred background of trees, a distant building, and a cloudy sky. The lighting suggests it's either dusk or dawn.
* **Motion:** The image captures the car in motion, with a slight motion blur to convey speed.
**Overall Impression:** The image conveys a sense of speed, luxury, and power. It's a well-composed shot that highlights the car's design and performance.
Do you want me to describe any specific aspect of the image in more detail, or perhaps analyze its composition?
```
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤️ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs **Phi-4-mini-instruct** using phi-4-mini-q4_0.gguf , llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limtations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Low | Very Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium Low | Low | CPU with more memory | Better accuracy while still being quantized |
| **Q8** | Medium | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
## **Included Files & Details**
### `google_gemma-3-4b-it-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `google_gemma-3-4b-it-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `google_gemma-3-4b-it-bf16-q8.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `google_gemma-3-4b-it-f16-q8.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `google_gemma-3-4b-it-q4_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `google_gemma-3-4b-it-q4_k_m.gguf`
- Similar to Q4_K.
- Another option for **low-memory CPU inference**.
### `google_gemma-3-4b-it-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `google_gemma-3-4b-it-q6_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `google_gemma-3-4b-it-q6_k_m.gguf`
- A mid-range **Q6_K** quantized model for balanced performance .
- Suitable for **CPU-based inference** with **moderate memory**.
### `google_gemma-3-4b-it-q8.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
# Gemma 3 model card
**Model Page**: [Gemma](https://ai.google.dev/gemma/docs/core)
**Resources and Technical Documentation**:
* [Gemma 3 Technical Report][g3-tech-report]
* [Responsible Generative AI Toolkit][rai-toolkit]
* [Gemma on Kaggle][kaggle-gemma]
* [Gemma on Vertex Model Garden][vertex-mg-gemma3]
**Terms of Use**: [Terms][terms]
**Authors**: Google DeepMind
## Model Information
Summary description and brief definition of inputs and outputs.
### Description
Gemma is a family of lightweight, state-of-the-art open models from Google,
built from the same research and technology used to create the Gemini models.
Gemma 3 models are multimodal, handling text and image input and generating text
output, with open weights for both pre-trained variants and instruction-tuned
variants. Gemma 3 has a large, 128K context window, multilingual support in over
140 languages, and is available in more sizes than previous versions. Gemma 3
models are well-suited for a variety of text generation and image understanding
tasks, including question answering, summarization, and reasoning. Their
relatively small size makes it possible to deploy them in environments with
limited resources such as laptops, desktops or your own cloud infrastructure,
democratizing access to state of the art AI models and helping foster innovation
for everyone.
### Inputs and outputs
- **Input:**
- Text string, such as a question, a prompt, or a document to be summarized
- Images, normalized to 896 x 896 resolution and encoded to 256 tokens
each
- Total input context of 128K tokens for the 4B, 12B, and 27B sizes, and
32K tokens for the 1B size
- **Output:**
- Generated text in response to the input, such as an answer to a
question, analysis of image content, or a summary of a document
- Total output context of 8192 tokens
## Credits
Thanks [Bartowski](https://huggingface.co/bartowski) for imartix upload. And your guidance on quantization that has enabled me to produce these gguf file. |
Mungert/Phi-4-mini-instruct.gguf | Mungert | 2025-06-15T19:39:04Z | 4,816 | 25 | null | [
"gguf",
"Demo",
"Phi 4 Mini",
"multilingual",
"reasoning",
"code generation",
"function calling",
"chat completion",
"memory efficient",
"low latency",
"128k context",
"text-generation",
"base_model:microsoft/Phi-4-mini-instruct",
"base_model:quantized:microsoft/Phi-4-mini-instruct",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-02-28T01:14:03Z | ---
license: mit
base_model:
- microsoft/Phi-4-mini-instruct
pipeline_tag: text-generation
tags:
- Demo
- Phi 4 Mini
- multilingual
- reasoning
- code generation
- function calling
- chat completion
- memory efficient
- low latency
- 128k context
---
# Model Summary
Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data. The model belongs to the Phi-4 model family and supports 128K token context length. The model underwent an enhancement process, incorporating both supervised fine-tuning and direct preference optimization to support precise instruction adherence and robust safety measures.
📰 [Phi-4-mini Microsoft Blog](https://aka.ms/phi4-feb2025) <br>
📖 [Phi-4-mini Technical Report](https://aka.ms/phi-4-multimodal/techreport) <br>
👩🍳 [Phi Cookbook](https://github.com/microsoft/PhiCookBook) <br>
🏡 [Phi Portal](https://azure.microsoft.com/en-us/products/phi) <br>
🖥️ Try It [Azure](https://aka.ms/phi-4-mini/azure), [Huggingface](https://huggingface.co/spaces/microsoft/phi-4-mini) <br>
**Phi-4**:
[[mini-instruct](https://huggingface.co/microsoft/Phi-4-mini-instruct) | [onnx](https://huggingface.co/microsoft/Phi-4-mini-instruct-onnx)];
[multimodal-instruct](https://huggingface.co/microsoft/Phi-4-multimodal-instruct); [gguf](https://huggingface.co/Mungert/Phi-4-mini-instruct.gguf)
## Usage
### Chat format
This format is used for general conversation and instructions:
```yaml
<|system|>Insert System Message<|end|><|user|>Insert User Message<|end|><|assistant|>
```
### Tool-Enabled Function-Calling Format
This format is used when the user wants the model to provide function calls based on the given tools. The user should define the available tools in the system prompt, wrapped by `<|tool|>` and `<|/tool|>` tokens. The tools must be specified in JSON format using a structured JSON dump.
```plaintext
<|system|>
You are a helpful assistant with some tools.
<|tool|>
[
{
"name": "get_weather_updates",
"description": "Fetches weather updates for a given city using the RapidAPI Weather API.",
"parameters": {
"city": {
"description": "The name of the city for which to retrieve weather information.",
"type": "str",
"default": "London"
}
}
}
]
<|/tool|>
<|end|>
<|user|>
What is the weather like in Paris today?
<|end|>
<|assistant|>
```
---
# <span style="color: #FF7F7F;">Unsloth Bug Fixes for Better Performance</span>
***Update (March 1, 2025)***
Applying all [Unsloth](https://huggingface.co/unsloth) fixes improved inference stability.
| # | Fix | Reason for Fix |
|----|--------------------------------------------------|------------------------------------------------|
| 1 | Changed the padding tag | The old padding tag could cause training issues. |
| 2 | Removed `{% else %}{{ eos_token }}` from chat template | Prevented extra EOS tokens that could degrade inference performance. |
| 3 | Replaced EOS with <\|end\|> | Avoided potential inference glitches. |
| 4 | Changed `unk_token` from EOS to `�` | Stopped unknown tokens from breaking inference. |
# <span id="testllm" style="color: #7F7FFF;">🚀 Phi 4 Mini Function Calling Test!</span>
If you have a minute, I’d really appreciate it if you could test my Phi-4-Mini-Instruct Demo at 👉 [Quantum Network Monitor](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Then toggle between the LLM Types Phi-4-Mini-Instruct is called TestLLM : TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs **Phi-4-mini-instruct** using phi-4-mini-q4_0.gguf , llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# <span style="color: #7FFF7F;">Phi-4-mini-instruct GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limtations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Low | Very Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium Low | Low | CPU with more memory | Better accuracy while still being quantized |
| **Q8** | Medium | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
## **Included Files & Details**
### `phi-4-mini-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `phi-4-mini-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `phi-4-mini-bf16-q8.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `phi-4-mini-f16-q8.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `phi-4-mini-q4_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `phi-4-mini-q4_k_m.gguf`
- Similar to Q4_K.
- Another option for **low-memory CPU inference**.
### `phi-4-mini-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `phi-4-mini-q6_k_l.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `phi-4-mini-q6_k_m.gguf`
- A mid-range **Q6_K** quantized model for balanced performance .
- Suitable for **CPU-based inference** with **moderate memory**.
### `phi-4-mini-q8.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
## Credits
Thanks [Bartowski](https://huggingface.co/bartowski) for imartix upload. And your guidance on quantization that has enabled me to produce these gguf file.
Thanks [Unsloth](https://huggingface.co/unsloth) for bug fixing many models.
Thanks for your support! 🙌 |
Peacemann/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B_LMUL | Peacemann | 2025-06-15T19:38:55Z | 0 | 0 | null | [
"qwen3",
"text-generation",
"lmul",
"research",
"experimental",
"conversational",
"base_model:deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
"base_model:finetune:deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
"license:mit",
"region:us"
] | text-generation | 2025-06-15T19:34:33Z | ---
license: mit
base_model: deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
tags:
- text-generation
- lmul
- research
- experimental
- qwen3
---
# L-Mul Optimized: deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
This is a modified version of DeepSeek AI's [DeepSeek-R1-0528-Qwen3-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B) model. The modification consists of replacing the standard attention mechanism with one that uses a custom, approximate matrix multiplication algorithm termed "L-Mul".
This work was performed as part of a research project to evaluate the performance and accuracy trade-offs of algorithmic substitutions in transformer architectures.
**This model is intended strictly for educational and scientific purposes.**
## Model Description
The core architecture of `deepseek-ai/DeepSeek-R1-0528-Qwen3-8B` is preserved. However, the standard `Qwen3Attention` modules have been dynamically replaced with a custom version that utilizes the `l_mul_attention` function for its core computations. This function is defined in the `lmul.py` file included in this repository.
- **Base Model:** [deepseek-ai/DeepSeek-R1-0528-Qwen3-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B)
- **Modification:** Replacement of standard attention with L-Mul approximate attention.
- **Primary Use-Case:** Research and educational analysis of algorithmic impact on LLMs.
## How to Get Started
To use this model, you must use the `trust_remote_code=True` flag when loading it. This is required to execute the custom `lmul.py` file that defines the new attention mechanism.
You can load the model directly from this repository using the `transformers` library:
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Define the repository ID for the specific model
repo_id = "Peacemann/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B_LMUL" # Replace with the correct repo ID if different
# Load the tokenizer and model, trusting the remote code to load lmul.py
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(
repo_id,
trust_remote_code=True,
torch_dtype=torch.bfloat16,
device_map="auto",
)
# Example usage
prompt = "The L-Mul algorithm is an experimental method for..."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
For high-throughput inference, you can use `vLLM`:
```python
from vllm import LLM
repo_id = "Peacemann/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B_LMUL" # Replace with the correct repo ID
llm = LLM(model=repo_id, trust_remote_code=True)
```
## Intended Uses & Limitations
This model is intended for researchers and students exploring the internal workings of LLMs. It is a tool for visualizing and analyzing the effects of fundamental algorithmic changes.
**This model is NOT intended for any commercial or production application.**
The modification is experimental. The impact on the model's performance, safety alignment, accuracy, and potential for generating biased or harmful content is **unknown and untested**. It inherits all limitations and biases of the original `DeepSeek-R1-0528-Qwen3-8B` model, and its behavior may be altered in unpredictable ways.
## Licensing Information
The use of this model is subject to the original **MIT License**. By using this model, you agree to the terms outlined in the license. The license can be found on the base model's Hugging Face page. |
Mungert/Seed-Coder-8B-Reasoning-GGUF | Mungert | 2025-06-15T19:38:52Z | 1,113 | 1 | transformers | [
"transformers",
"gguf",
"text-generation",
"arxiv:2506.03524",
"base_model:ByteDance-Seed/Seed-Coder-8B-Base",
"base_model:quantized:ByteDance-Seed/Seed-Coder-8B-Base",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-06-08T17:27:28Z | ---
library_name: transformers
pipeline_tag: text-generation
license: mit
base_model:
- ByteDance-Seed/Seed-Coder-8B-Base
---
# <span style="color: #7FFF7F;">Seed-Coder-8B-Reasoning GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`5787b5da`](https://github.com/ggerganov/llama.cpp/commit/5787b5da57e54dba760c2deeac1edf892e8fc450).
## <span style="color: #7FFF7F;"> Quantization beyond the IMatrix</span>
Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.
I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
This does create larger model files but increases precision for a given model size.
### **Please provide feedback on how you find this method performs**
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Hybrid Precision Models (e.g., `bf16_q8_0`, `f16_q4_K`) – Best of Both Worlds**
These formats selectively **quantize non-essential layers** while keeping **key layers in full precision** (e.g., attention and output layers).
- Named like `bf16_q8_0` (meaning **full-precision BF16 core layers + quantized Q8_0 other layers**).
- Strike a **balance between memory efficiency and accuracy**, improving over fully quantized models without requiring the full memory of BF16/F16.
📌 **Use Hybrid Models if:**
✔ You need **better accuracy than quant-only models** but can’t afford full BF16/F16 everywhere.
✔ Your device supports **mixed-precision inference**.
✔ You want to **optimize trade-offs** for production-grade models on constrained hardware.
📌 **Avoid Hybrid Models if:**
❌ Your target device doesn’t support **mixed or full-precision acceleration**.
❌ You are operating under **ultra-strict memory limits** (in which case use fully quantized formats).
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **very high memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **very high memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
### **Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)**
- *Ultra-low-bit quantization (1 2-bit) with **extreme memory efficiency**.
- **Use case**: Best for cases were you have to fit the model into very constrained memory
- **Trade-off**: Very Low Accuracy. May not function as expected. Please test fully before using.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------------------|------------------|------------------|----------------------------------|--------------------------------------------------------------|
| **BF16** | Very High | High | BF16-supported GPU/CPU | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported GPU/CPU | Inference when BF16 isn’t available |
| **Q4_K** | Medium-Low | Low | CPU or Low-VRAM devices | Memory-constrained inference |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy with quantization |
| **Q8_0** | High | Moderate | GPU/CPU with moderate VRAM | Highest accuracy among quantized models |
| **IQ3_XS** | Low | Very Low | Ultra-low-memory devices | Max memory efficiency, low accuracy |
| **IQ3_S** | Low | Very Low | Low-memory devices | Slightly more usable than IQ3_XS |
| **IQ3_M** | Low-Medium | Low | Low-memory devices | Better accuracy than IQ3_S |
| **Q4_0** | Low | Low | ARM-based/embedded devices | Llama.cpp automatically optimizes for ARM inference |
| **Ultra Low-Bit (IQ1/2_*)** | Very Low | Extremely Low | Tiny edge/embedded devices | Fit models in extremely tight memory; low accuracy |
| **Hybrid (e.g., `bf16_q8_0`)** | Medium–High | Medium | Mixed-precision capable hardware | Balanced performance and memory, near-FP accuracy in critical layers |
---
# Seed-Coder-8B-Reasoning
<div align="left" style="line-height: 1;">
<a href="https://bytedance-seed-coder.github.io/" target="_blank" style="margin: 2px;">
<img alt="Homepage" src="https://img.shields.io/badge/Seed--Coder-Homepage-a468fe?color=a468fe&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://arxiv.org/abs/2506.03524" target="_blank" style="margin: 2px;">
<img alt="Technical Report" src="https://img.shields.io/badge/arXiv-Technical%20Report-brightgreen?logo=arxiv&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://huggingface.co/ByteDance-Seed" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-ByteDance%20Seed-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://github.com/ByteDance-Seed/Seed-Coder/blob/master/LICENSE" style="margin: 2px;">
<img alt="License" src="https://img.shields.io/badge/License-MIT-f5de53?color=f5de53&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
## Introduction
We are thrilled to introduce Seed-Coder, a powerful, transparent, and parameter-efficient family of open-source code models at the 8B scale, featuring base, instruct, and reasoning variants. Seed-Coder contributes to promote the evolution of open code models through the following highlights.
- **Model-centric:** Seed-Coder predominantly leverages LLMs instead of hand-crafted rules for code data filtering, minimizing manual effort in pretraining data construction.
- **Transparent:** We openly share detailed insights into our model-centric data pipeline, including methods for curating GitHub data, commits data, and code-related web data.
- **Powerful:** Seed-Coder achieves state-of-the-art performance among open-source models of comparable size across a diverse range of coding tasks.
<p align="center">
<img width="100%" src="imgs/seed-coder_intro_performance.png">
</p>
This repo contains the **Seed-Coder-8B-Reasoning** model, which has the following features:
- Type: Causal language models
- Training Stage: Pretraining & Post-training
- Data Source: Public datasets
- Context Length: 65,536
## Model Downloads
| Model Name | Length | Download | Notes |
|---------------------------------------------------------|-----------|------------------------------------|-----------------------|
| Seed-Coder-8B-Base | 32K | 🤗 [Model](https://huggingface.co/ByteDance-Seed/Seed-Coder-8B-Base) | Pretrained on our model-centric code data. |
| Seed-Coder-8B-Instruct | 32K | 🤗 [Model](https://huggingface.co/ByteDance-Seed/Seed-Coder-8B-Instruct) | Instruction-tuned for alignment with user intent. |
| 👉 **Seed-Coder-8B-Reasoning** | 64K | 🤗 [Model](https://huggingface.co/ByteDance-Seed/Seed-Coder-8B-Reasoning) | RL trained to boost reasoning capabilities. |
| Seed-Coder-8B-Reasoning-bf16 | 64K | 🤗 [Model](https://huggingface.co/ByteDance-Seed/Seed-Coder-8B-Reasoning-bf16) | RL trained to boost reasoning capabilities. |
## Requirements
You will need to install the latest versions of `transformers` and `accelerate`:
```bash
pip install -U transformers accelerate
```
## Quickstart
Here is a simple example demonstrating how to load the model and perform code generation using the Hugging Face `pipeline` API:
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "ByteDance-Seed/Seed-Coder-8B-Reasoning"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)
messages = [
{"role": "user", "content": "Write a quick sort algorithm."},
]
input_ids = tokenizer.apply_chat_template(
messages,
tokenize=True,
return_tensors="pt",
add_generation_prompt=True,
).to(model.device)
outputs = model.generate(input_ids, max_new_tokens=16384)
response = tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True)
print(response)
```
## Evaluation
Seed-Coder-8B-Reasoning strikes impressive performance on competitive programming, demonstrating that smaller LLMs can also be competent on complex reasoning tasks. Our model surpasses QwQ-32B and DeepSeek-R1 on IOI'2024, and achieves an ELO rating comparable to o1-mini on Codeforces contests.
<div style="display: flex; justify-content: center;">
<img src="imgs/reasoning-ioi.jpg" width="61%" />
<img src="imgs/reasoning-codeforces.jpg" width="39%" />
</div>
For detailed benchmark performance, please refer to our [📑 Technical Report](https://github.com/ByteDance-Seed/Seed-Coder/blob/master/Seed-Coder.pdf).
## License
This project is licensed under the MIT License. See the [LICENSE file](https://github.com/ByteDance-Seed/Seed-Coder/blob/master/LICENSE) for details.
## Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{seed2025seedcoderletcodemodel,
title={{Seed-Coder}: Let the Code Model Curate Data for Itself},
author={{ByteDance Seed} and Yuyu Zhang and Jing Su and Yifan Sun and Chenguang Xi and Xia Xiao and Shen Zheng and Anxiang Zhang and Kaibo Liu and Daoguang Zan and Tao Sun and Jinhua Zhu and Shulin Xin and Dong Huang and Yetao Bai and Lixin Dong and Chao Li and Jianchong Chen and Hanzhi Zhou and Yifan Huang and Guanghan Ning and Xierui Song and Jiaze Chen and Siyao Liu and Kai Shen and Liang Xiang and Yonghui Wu},
year={2025},
eprint={2506.03524},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2506.03524},
}
```
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
Mungert/lucid-v1-nemo-GGUF | Mungert | 2025-06-15T19:38:49Z | 1,280 | 1 | transformers | [
"transformers",
"gguf",
"axolotl",
"unsloth",
"torchtune",
"en",
"base_model:mistralai/Mistral-Nemo-Base-2407",
"base_model:quantized:mistralai/Mistral-Nemo-Base-2407",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | 2025-06-08T14:12:44Z | ---
library_name: transformers
base_model:
- mistralai/Mistral-Nemo-Base-2407
language:
- en
tags:
- axolotl
- unsloth
- torchtune
license: other
---
# <span style="color: #7FFF7F;">lucid-v1-nemo GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`5787b5da`](https://github.com/ggerganov/llama.cpp/commit/5787b5da57e54dba760c2deeac1edf892e8fc450).
## <span style="color: #7FFF7F;"> Quantization beyond the IMatrix</span>
Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.
I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
This does create larger model files but increases precision for a given model size.
### **Please provide feedback on how you find this method performs**
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Hybrid Precision Models (e.g., `bf16_q8_0`, `f16_q4_K`) – Best of Both Worlds**
These formats selectively **quantize non-essential layers** while keeping **key layers in full precision** (e.g., attention and output layers).
- Named like `bf16_q8_0` (meaning **full-precision BF16 core layers + quantized Q8_0 other layers**).
- Strike a **balance between memory efficiency and accuracy**, improving over fully quantized models without requiring the full memory of BF16/F16.
📌 **Use Hybrid Models if:**
✔ You need **better accuracy than quant-only models** but can’t afford full BF16/F16 everywhere.
✔ Your device supports **mixed-precision inference**.
✔ You want to **optimize trade-offs** for production-grade models on constrained hardware.
📌 **Avoid Hybrid Models if:**
❌ Your target device doesn’t support **mixed or full-precision acceleration**.
❌ You are operating under **ultra-strict memory limits** (in which case use fully quantized formats).
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **very high memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **very high memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
### **Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)**
- *Ultra-low-bit quantization (1 2-bit) with **extreme memory efficiency**.
- **Use case**: Best for cases were you have to fit the model into very constrained memory
- **Trade-off**: Very Low Accuracy. May not function as expected. Please test fully before using.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------------------|------------------|------------------|----------------------------------|--------------------------------------------------------------|
| **BF16** | Very High | High | BF16-supported GPU/CPU | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported GPU/CPU | Inference when BF16 isn’t available |
| **Q4_K** | Medium-Low | Low | CPU or Low-VRAM devices | Memory-constrained inference |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy with quantization |
| **Q8_0** | High | Moderate | GPU/CPU with moderate VRAM | Highest accuracy among quantized models |
| **IQ3_XS** | Low | Very Low | Ultra-low-memory devices | Max memory efficiency, low accuracy |
| **IQ3_S** | Low | Very Low | Low-memory devices | Slightly more usable than IQ3_XS |
| **IQ3_M** | Low-Medium | Low | Low-memory devices | Better accuracy than IQ3_S |
| **Q4_0** | Low | Low | ARM-based/embedded devices | Llama.cpp automatically optimizes for ARM inference |
| **Ultra Low-Bit (IQ1/2_*)** | Very Low | Extremely Low | Tiny edge/embedded devices | Fit models in extremely tight memory; low accuracy |
| **Hybrid (e.g., `bf16_q8_0`)** | Medium–High | Medium | Mixed-precision capable hardware | Balanced performance and memory, near-FP accuracy in critical layers |
---
Lucid is a family of AI language models developed and used by DreamGen with a **focus on role-play and story-writing capabilities**.
<a href="https://dreamgen.com" style="display: flex; justify-content: center;">
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/banner.webp" style="border-radius: 1rem; max-width: 90vw; max-height: 15rem; margin: 0;"/>
</a>
<a href="https://dreamgen.com/discord" style="display: flex; justify-content: center;">
<strong>Join our Discord</strong>
</a>
## Table of Contents
- [Why Lucid](#why-lucid)
- [Prompt Template](#prompt-template)
- [Role-Play & Story-Writing](#role-play--story-writing)
- [System Prompt](#system-prompt)
- [Intro](#intro)
- [Roles](#roles)
- [Structured Scenario Information](#structured-scenario-information)
- [Unstructured Scenario Instructions](#unstructured-scenario-instructions)
- [Controlling Response Length](#controlling-response-length)
- [Instructions (OOC)](#instructions-ooc)
- [Controlling Plot Development](#controlling-plot-development)
- [Writing Assistance](#writing-assistance)
- [Planning The Next Scene](#planning-the-next-scene)
- [Suggesting New Characters](#suggesting-new-characters)
- [Sampling Params](#sampling-params)
- [SillyTavern](#sillytavern)
- [Role-Play](#role-play-preset)
- [Story-Writing](#story-writing-preset)
- [Assistant](#assistant-preset)
- [API: Local](#using-local-api)
- [API: DreamGen (free)](#using-dreamgen-api-free)
- [DreamGen Web App (free)](#dreamgen-web-app-free)
- [Model Training Details](#model-training-details)
## Why Lucid
Is Lucid the right model for you? I don't know. But here is what I do know:
- Focused on role-play & story-writing.
- Suitable for all kinds of writers and role-play enjoyers:
- For world-builders who want to specify every detail in advance: plot, setting, writing style, characters, locations, items, lore, etc.
- For intuitive writers who start with a loose prompt and shape the narrative through instructions (OCC) as the story / role-play unfolds.
- Support for **multi-character role-plays**:
- Model can automatically pick between characters.
- Support for **inline writing instructions (OOC)**:
- Controlling plot development (say what should happen, what the characters should do, etc.)
- Controlling pacing.
- etc.
- Support for **inline writing assistance**:
- Planning the next scene / the next chapter / story.
- Suggesting new characters.
- etc.
- Support for **reasoning (opt-in)**.
- Detailed README.md with documentation, examples and configs.
Sounds interesting? Read on!
## Prompt Template
The Lucid prompt template is an extension of the Llama 3 template. These are the extensions:
**Writing**
- The Lucid prompt template adds a new `writer` role.
- While story-writing, the `writer` message header looks like this: `<|start_header_id|>writer<|end_header_id|>`.
- While role-playing, the `writer` message header differs based on whether it's a message from a character or the narrator:
- For messages from characters: `<|start_header_id|>writer character {{character_name}}<|end_header_id|>` where `{{character_name}}` would be the actual character name.
- For messages from the narrator: `<|start_header_id|>writer narrative<|end_header_id|>`.
- The model is trained to predict the writer header, that means **you can have it generate multiple messages at once** and it will naturally pick between the characters and narrative.
Some things to keep in mind:
- You may have a sequence of roles like this: `system` -> `writer character Jon` -> `writer character Amy` -> `user` -> `assistant` -> `writer narration` -> ...
- You should use `writer character Jon Doe` when sending a message for a particular character (in this case Jon Doe), **including the user character**.
- You should use `user` when providing some instructions (OOC) to the model.
**Reasoning**
- The Lucid prompt template adds 2 new tokens for reasoning: `<|reasoning_start|>` and `<|reasoning_end|>`.
- The **reasoning is opt-in**, you should put `Think deeply before answering.` (or similar) into your system prompt or in your user prompt to enable it.
### Assistant: Simple
```
<|start_header_id|>system<|end_header_id|>
You are a reliable, general-purpose AI assistant.<|eot_id|><|start_header_id|>user<|end_header_id|>
Why is the sky blue?<|eot_id|><|start_header_id|>assistant<|end_header_id|>
The sky appears blue due to a phenomenon called **Rayleigh scattering**, a type of light scattering caused by tiny particles in the atmosphere. Here's a simplified breakdown:
...
```
### Assistant: Reasoning
Note that reasoning is mostly intended for assistant tasks (knowledge, math, coding, etc.) and much less for writing. The smaller Lucid models (<=12B) can produce plausible reasoning traces, but reasoning does not improve their performance as much as it does for larger models.
```
<|start_header_id|>system<|end_header_id|>
You are a reliable, general-purpose AI assistant.<|eot_id|><|start_header_id|>user<|end_header_id|>
Why is the sky blue?
Think deeply before answering.<|eot_id|><|start_header_id|>assistant<|end_header_id|>
<|reasoning_start|>Okay, so I need to figure out why the sky is blue. Let me start by recalling what I know...
...<|reasoning_end|>The sky appears blue due to a phenomenon called **Rayleigh scattering**. Here's a concise breakdown:
...
```
## Role-Play & Story-Writing
Quick summary:
- `system` is used for setting up the scenario for the story / role-play (e.g. plot, setting, writing style, character descriptions, etc.)
- `writer` is used for the contents of the story / role-play.
- `user` is used for instructions (OOC), e.g. how the story should develop, etc.
- `assistant` is used for providing writing assistance, e.g. planning the next scene, suggesting new characters, etc.
Some things to keep in mind:
- You may have a sequence of roles like this: `system` -> `writer character Jon` -> `writer character Amy` -> `user` -> `assistant` -> `writer narration` -> ...
- You should use `writer character Jon Doe` when sending a message for a particular character (in this case Jon Doe), **including the user character**.
- You should use `user` when providing some instructions (OOC) to the model.
### System Prompt
The model is very flexible in terms of how you can structure the system prompt.
Here are some recommendations and examples.
#### Intro
It's recommended to start the system prompt with an intro like this:
```
You are a skilled {{writing_kind}} writer and writing assistant with expertise across all genres{{genre_suffix}}.
```
- `{{writing_kind}}` should be `fiction` for story-writing and `role-play` for role-playing.
- `{{genre_suffix}}` can be empty, or something like ` (but you specialize in horror)`.
#### Roles
It's recommended to explain the writing roles in the system prompt.
For story-writing, you can use the following:
```
You will perform several tasks, switching roles as needed:
- Writing: Use the `writer` role to write a story based on the provided information and user instructions.
- Other: Use the `assistant` role for any other tasks the user may request.
```
For role-playing, you can use the following:
```
You will perform several tasks, switching roles as needed:
- Role-playing: Use the `writer` role to write a role-play based on the provided information and user instructions. Use the `writer character <character_name>` role for dialog or when acting as a specific character, use the `writer narrative` role for narration.
- Other: Use the `assistant` role for any other tasks the user may request.
```
#### Structured Scenario Information
The model understands structured scenario information, consisting of plot, style, setting, characters, etc. (this is not an exhaustive list, and all sections are optional).
<details>
<summary><strong>Markdown Style</strong></summary>
```
# Role-Play Information
## Plot
{{ description of the plot, can be open-ended }}
## Previous Events
{{ summary of what happened before -- like previous chapters or scenes }}
## Setting
{{ description of the setting }}
## Writing Style
{{ description of the writing style }}
## Characters
## {{character's name}}
{{ description of the character }}
## {{character's name}}
{{ description of the character }}
## Locations
{{ like characters }}
## Items
{{ like characters }}
```
</details>
<details>
<summary><strong>XML Style</strong></summary>
```
# Role-Play Information
<plot>
{{ description of the plot, can be open-ended }}
</plot>
<previous_events>
{{ summary of what happened before -- like previous chapters or scenes }}
</previous_events>
<setting>
{{ description of the setting }}
</setting>
<writing_style>
{{ description of the writing style }}
</writing_style>
<characters>
<character>
<name>{{ character's name }}</name>
{{ description of the character }}
</character>
<character>
<name>{{ character's name }}</name>
{{ description of the character }}
</character>
</characters>
<locations>
{{ like characters }}
</locations>
<items>
{{ like characters }}
</items>
```
</details>
Other recommendations:
- Write descriptions as plain sentences / paragraphs or bullet points.
- If the plot description is open-ended, put something like this at the start:
```
[ The following plot description is open-ended and covers only the beginning of the {{writing_kind}}. ]
```
Where `{{writing_kind}}` is `story` or `role-play`.
**Example: Science Fiction Story**
<details>
<summary>Input</summary>
```
<|start_header_id|>system<|end_header_id|>
You are a skilled fiction writer and writing assistant with expertise across all genres (but you specialize in Science Fiction).
You will perform several tasks, switching roles as needed:
- Writing: Use the `writer` role to write a story based on the provided information and user instructions.
- Other: Use the `assistant` role for any other tasks the user may request.
# Story Information
## Plot
[ The following plot description is open-ended and covers only the beginning of the story. ]
In the bleak expanse of space, Tess, a seasoned scavenger clad in a voidskin, meticulously prepares to breach a derelict ark vessel. Her partner, the pragmatic Lark, pilots their ship, guiding Tess towards a promising entry point on the starboard side. This particular spot, identified by its unusual heat signature, hints at a potentially intact nerve center, a relic of the ship's former life. As they coordinate their approach, fragments of their shared past surface: their daring escape from the dismantling of Nova-Tau Hub and the acquisition of their vessel, the "Rustbucket," burdened by a substantial debt. The lure of riches within the unbreached ark vessel hangs heavy in the air, tempered by the knowledge that other scavengers will soon descend, transforming the find into a frenzied feeding ground. The need for speed and precision is paramount as Tess and Lark initiate the breach, stepping into the unknown depths of the derelict vessel.
## Setting
The story unfolds in a distant, unspecified future, long after humanity has colonized space. Earth, or "the homeland," is a distant memory, its oceans and creatures relegated to historical data. The primary setting is the cold, dark void of space, where derelict "ark vessels"—massive vessels designed to carry generations of colonists—drift as "spacegrave," decaying carcasses ripe for scavenging.
Physically, the story takes place within and around these ark vessels and the scavengers' smaller, utilitarian "Rustbucket." The ark vessels are described as station-sized, containing server farms, cryotube chambers filled with broken glass, and dimly lit corridors marked with pictographic signs. Colors are muted, dominated by blacks, blues, and the occasional "blood-red hotspot" indicating active electronics. The air is thin, oxygen scarce, and the temperature varies from frigid to warm near active systems. Sounds include the hum of generators, the hiss of airlocks, and the clank of magnetized boots on metal hulls.
Socially, the setting is defined by a corpostate hegemony, leaving many, like Tess and Lark, deeply in debt and forced into dangerous scavenging work. The culture is pragmatic and utilitarian, valuing raw materials over historical preservation. Technology is advanced, featuring jump drives, voidskins, slicers (bladed tools), and cryotubes, but is often unreliable and jury-rigged. Legal claims to derelict ships are determined by the presence of living survivors, creating a moral conflict for the scavengers.
## Writing Style
- The story employs a first-person perspective, narrated by Tess, offering a limited yet intimate view of events. Tess's voice is informal, pragmatic, and tinged with cynicism ("Rustbucket"), reflecting her working-class background and life as a scavenger. Her reliability as a narrator is questionable, as she often rationalizes morally ambiguous decisions.
- The tense is primarily present, creating immediacy, with occasional shifts to the past for reflection.
- The tone blends cynical detachment with underlying sentimentality, revealing a capacity for empathy beneath Tess's tough exterior.
- Vivid imagery and sensory details paint a bleak yet compelling picture of decaying ark vessels and the harsh realities of space scavenging.
- Simple, straightforward language is interspersed with technical jargon, grounding the story in its sci-fi setting.
- Figurative language, such as the central "spacegrave" metaphor, enhances imagery and conveys complex ideas about decay and lost potential.
- The dialogue is authentic, advancing the plot and revealing character traits. Tess's pragmatic speech contrasts with Lark's more intellectual and idealistic language, highlighting their differing personalities.
- The story explores themes of survival, loyalty, and the conflict between pragmatism and morality, reinforced by recurring motifs and stylistic choices.
## Characters
### Tess
Tess is a hardened scavenger in her late twenties or early thirties with a practical approach to surviving in the harsh void of space. Years of exostation welding have made her physically strong, capable of handling heavy equipment and enduring difficult conditions. She carries psychological scars from past violence during her time at Nova-Tau, which manifests as emotional guardedness. Though her exterior is stoic and her communication style blunt, she harbors deep loyalty toward her partner Lark. Tess often dismisses speculation and sentiment in favor of immediate action, yet occasionally allows herself brief moments of peace and reflection. Her speech is terse and direct, complementing her no-nonsense approach to their dangerous profession. Beneath her pragmatic facade lies an ongoing internal struggle between the ruthlessness required for survival and her persistent humanity.
### Lark
Lark serves as Tess's partner and counterbalance, bringing curiosity and emotional expressiveness to their scavenging team. Approximately Tess's age, Lark possesses sharp analytical skills that prove invaluable during their operations. She excels at technical scanning and data interpretation, often identifying optimal entry points and valuable salvage opportunities. Unlike Tess, Lark maintains an enthusiastic interest in history and cultural artifacts, collecting media from different eras and frequently sharing historical knowledge. Her communication style is conversational and expressive, often employing humor to process difficult situations. While practical about their survival needs, Lark struggles more visibly with the moral implications of their work. She values knowledge and understanding alongside material gain, seeing connections between past civilizations and their present circumstances. Her relationship with Tess forms the emotional core of her life, built on shared experiences and mutual dependence.
## Locations
### Ark Vessel
- Massive, derelict spacecraft, heavily decayed from centuries adrift in space.
- Originally built to sustain multiple generations during interstellar travel.
- Contains a variety of sections, including decayed living quarters, server farms filled with obsolete technology, cryotube chambers, vast storage facilities, and the potential for long-dead gardens or greenspaces.
- The interior is a hazardous environment due to structural instability and the potential presence of surviving inhabitants or automated defense systems.
- A prime target for scavengers seeking valuable resources, historical artifacts, and exotic technology.
### The Rustbucket
- A small, old, and unreliable spacecraft, stolen from Nova-Tau Hub.
- Equipped with a jump drive (the source of a significant debt) for interstellar travel.
- Contains an airlock for accessing external environments and basic, often strained, life support systems, including oxygen scrubbers.
- The cramped interior serves as both living quarters and workspace, filled with salvage equipment, tools, and essential living supplies.
## Items
### Voidskin
Protective spacesuit with sealed atmosphere, magnetic boots, and night vision. Essential for surviving vacuum, temperature extremes, and radiation. Re-zippable design.
### Slicer
Cutting tool/weapon projecting an energy blade. Used to slice through durable materials like spaceship hulls.
### Jump Drive
Enables faster-than-light interstellar travel. Crucial for trade, exploration, and scavenging. Expensive technology.
## Story Info
- Length: 2000 words.
- Quality: 10 / 10 (Very Good)<|eot_id|>
```
</details>
<details>
<summary>Output (partial)</summary>
```
<|start_header_id|>writer<|end_header_id|>
My voidskin is sealed, my slicer primed and ready. My magnetized boots clank against the rusted hull of the ark vessel we’ve targeted for this run. It’s the first of the trip, so we have to make it count. If we don’t come out with a good haul, we’re as good as dead. Our life-support systems are almost completely out, and the jump drive on the _Rustbucket_ is as old as the day is long.
“You’ve got the starboard side, right?” Lark’s voice comes through the comm, clear and crisp.
“Yeah, I’ve got it. I can see the hotspot, too,” I say. “It’s about a kilometer up, toward the nose.”
“That’s probably where you’ll find the nerve center. If we get lucky, it won’t be a fucking server farm.”
“Wouldn’t count on it, though.” My breath fogs in front of me, visible in the low light.
“Probably right.”
We’d gotten lucky with this ark. It was station-sized, which meant more space for salvage, and it hadn’t been breached yet. Its heat map showed a few blood-red hotspots, and there were no signs of salvage operations in the vicinity. We’d be first on site, which meant we’d have the pick of the litter, at least for a few hours. Then the rest of them would come.
I glance at my wrist screen and watch the counter. Thirty minutes before we’re supposed to rendezvous again.
I continue up the hull, slicer out and ready. The skin of the ark is so degraded it feels like it’ll crumble under my feet. I wonder how many centuries it’s been out here. How many more it has left before it disintegrates. We’re out far enough from the gravity wells that it doesn’t have a set orbit. Just adrift, spinning in space. A spacegrave.
Lark and I found the first spacegrave just after we’d gotten the _Rustbucket_. The name’s a little on the nose, but it’s what she’d named her when she was a kid, so I’d let her keep it. Even after the Nova-Tau hub dismantling, even after we stole her, she kept the name. That’s how much it meant to her. _Rustbucket_, a small ship with a big jump drive that was just big enough for two.
It was just the two of us. Two young women, alone on a small ship, far from home. We’d been lucky to get her at all. It wasn’t often that a small ship got fitted with a jump drive. They were too expensive, too difficult to maintain. Not many people could afford them. Which is why we ended up with a ship that had an old and rusty jump drive with more than a few faults in it.
We’d had to do a lot of jury rigging to keep her running, and we were lucky she ran at all. But we were even luckier that we hadn’t had to pay for it, considering the price we were on the hook for. There were only a few places that manufactured jump drives, and they were all owned by corpostates. Everything we did, everything we owned, was owed to them. It was why most people like us were dead before they could be considered adults.
We were different, though. We were scavengers. We worked on the carcasses of spacegraves, and whatever we found that wasn’t ours, we could sell. It helped us get by, helped us keep from starving out here in the void.
When we came across our first spacegrave, Lark had almost cried. I’d thought it was the best thing that could have happened to us. Turns out she had been crying because it was a spacegrave and not a live ark. Live arks were rarer than spacegraves. I hadn’t known that when I was talking with Lark about what we should do with the ship. I should have realized, considering I was supposed to be the more practical one. Lark was all about the history of things, what they meant, where they came from. Me, I just saw a carcass ripe for picking.
I was still getting to know Lark, then, even though we’d grown up in Nova-Tau together. We’d worked together, in the exostation welders, but that was all. We’d both been young. Too young to be welders, honestly, but too old to work anywhere else. There weren’t many places to work on an exostation. And you could only be a child for so long. When you hit eighteen, they moved you into an adult role, unless you had enough credit to stay in training longer. No one ever had enough credit for that.
It had been a strange and violent world, Nova-Tau. All of the exostations were. They all followed the same pattern. You were born into it, grew up in it, learned to survive it. Most people didn’t. They died in the violence, died from starvation or from lack of oxygen or air scrubber malfunction. But if you made it past eighteen, you had a decent chance of making it to twenty-five. From there, your chances rose exponentially. After twenty-five, you could even get off of it, if you wanted. At least, that’s what I’d heard.
I’d never met anyone who’d made it off of it. Well, I take that back. I’d met a lot of people who’d left Nova-Tau, but none of them had made it off. They’d died trying. Maybe Lark and I would be the exception to the rule.
I reach the hotspot and stop to examine it. It’s a small area, just a meter and a half in diameter. I hold up my wrist screen. It shows the heat of the area is concentrated, but not too intense. Not enough to burn through a voidskin. I use the slicer to cut out a small square, then press my hand against it. It’s warm, almost hot, but nothing that’ll burn. Good. If I couldn’t get in, it would be a waste of time and energy.
I grab the edges of the square and pull. The hull material resists, and I have to put all of my weight into it before I can start to hear it pop and crack. Finally, it gives way, and the hull around me shudders. I jump back, hoping it doesn’t collapse around me. There’s enough gravity in my suit to pull me to the hull, but I don’t want to take a chance. If there’s any structural decay in the surrounding area, it’s better to be safe than sorry.
Once the shaking subsides, I go back to the hole. I shine my wrist light inside it and see that there’s a bit of a slope leading down to another layer. I turn off my boots and go through.
The airlock is a familiar sight. I’ve been through so many, they’re all the same, now. Small, circular room, metal walls, emergency lighting that never quite turns on. There’s no light on this one, either. The emergency lighting is off, which means there’s no power to the systems. I’m surprised there’s power at all. Most of these older arks have long since run out of power. They’re just cold and dark inside.
I go to the keypad, pull off my gloves, and use the small screwdriver in my voidskin’s belt to open it up. Once I have it open, I can bypass the security protocols. I’ve done this so many times, it almost feels routine. I don’t even need the manual for it anymore.
“This one looks promising,” I say.
“How so?” Lark asks.
“The power is on. At least for the airlock. I might be able to get to the nerve center from here.”
“Good luck. I hope you can. If you find anything useful, we might actually have a chance.”
“Aren’t you always the optimist?” I ask with a smirk.
I wait for the airlock to depressurize. It takes longer than I want it to, but it eventually does. Once it does, I open the inner door and step inside. The ship smells old. Not like it hasn’t been used in years, but like it’s been here for centuries. It’s faint, but it’s there. I’m not sure if it’s possible for a place to smell old, but this one does.
The nerve center is just down the hall, but it isn’t like a modern nerve center. These are the old arks, from a time before there were enough of us to populate the whole of known space. Before we left Earth. Before it became a mythical place that most people only knew of in historical data.
```
</details>
<details>
<summary>Output (partial, formatted)</summary>
<blockquote>
My voidskin is sealed, my slicer primed and ready. My magnetized boots clank against the rusted hull of the ark vessel we’ve targeted for this run. It’s the first of the trip, so we have to make it count. If we don’t come out with a good haul, we’re as good as dead. Our life-support systems are almost completely out, and the jump drive on the _Rustbucket_ is as old as the day is long.
“You’ve got the starboard side, right?” Lark’s voice comes through the comm, clear and crisp.
“Yeah, I’ve got it. I can see the hotspot, too,” I say. “It’s about a kilometer up, toward the nose.”
“That’s probably where you’ll find the nerve center. If we get lucky, it won’t be a fucking server farm.”
“Wouldn’t count on it, though.” My breath fogs in front of me, visible in the low light.
“Probably right.”
We’d gotten lucky with this ark. It was station-sized, which meant more space for salvage, and it hadn’t been breached yet. Its heat map showed a few blood-red hotspots, and there were no signs of salvage operations in the vicinity. We’d be first on site, which meant we’d have the pick of the litter, at least for a few hours. Then the rest of them would come.
I glance at my wrist screen and watch the counter. Thirty minutes before we’re supposed to rendezvous again.
I continue up the hull, slicer out and ready. The skin of the ark is so degraded it feels like it’ll crumble under my feet. I wonder how many centuries it’s been out here. How many more it has left before it disintegrates. We’re out far enough from the gravity wells that it doesn’t have a set orbit. Just adrift, spinning in space. A spacegrave.
Lark and I found the first spacegrave just after we’d gotten the _Rustbucket_. The name’s a little on the nose, but it’s what she’d named her when she was a kid, so I’d let her keep it. Even after the Nova-Tau hub dismantling, even after we stole her, she kept the name. That’s how much it meant to her. _Rustbucket_, a small ship with a big jump drive that was just big enough for two.
It was just the two of us. Two young women, alone on a small ship, far from home. We’d been lucky to get her at all. It wasn’t often that a small ship got fitted with a jump drive. They were too expensive, too difficult to maintain. Not many people could afford them. Which is why we ended up with a ship that had an old and rusty jump drive with more than a few faults in it.
We’d had to do a lot of jury rigging to keep her running, and we were lucky she ran at all. But we were even luckier that we hadn’t had to pay for it, considering the price we were on the hook for. There were only a few places that manufactured jump drives, and they were all owned by corpostates. Everything we did, everything we owned, was owed to them. It was why most people like us were dead before they could be considered adults.
We were different, though. We were scavengers. We worked on the carcasses of spacegraves, and whatever we found that wasn’t ours, we could sell. It helped us get by, helped us keep from starving out here in the void.
When we came across our first spacegrave, Lark had almost cried. I’d thought it was the best thing that could have happened to us. Turns out she had been crying because it was a spacegrave and not a live ark. Live arks were rarer than spacegraves. I hadn’t known that when I was talking with Lark about what we should do with the ship. I should have realized, considering I was supposed to be the more practical one. Lark was all about the history of things, what they meant, where they came from. Me, I just saw a carcass ripe for picking.
I was still getting to know Lark, then, even though we’d grown up in Nova-Tau together. We’d worked together, in the exostation welders, but that was all. We’d both been young. Too young to be welders, honestly, but too old to work anywhere else. There weren’t many places to work on an exostation. And you could only be a child for so long. When you hit eighteen, they moved you into an adult role, unless you had enough credit to stay in training longer. No one ever had enough credit for that.
It had been a strange and violent world, Nova-Tau. All of the exostations were. They all followed the same pattern. You were born into it, grew up in it, learned to survive it. Most people didn’t. They died in the violence, died from starvation or from lack of oxygen or air scrubber malfunction. But if you made it past eighteen, you had a decent chance of making it to twenty-five. From there, your chances rose exponentially. After twenty-five, you could even get off of it, if you wanted. At least, that’s what I’d heard.
I’d never met anyone who’d made it off of it. Well, I take that back. I’d met a lot of people who’d left Nova-Tau, but none of them had made it off. They’d died trying. Maybe Lark and I would be the exception to the rule.
I reach the hotspot and stop to examine it. It’s a small area, just a meter and a half in diameter. I hold up my wrist screen. It shows the heat of the area is concentrated, but not too intense. Not enough to burn through a voidskin. I use the slicer to cut out a small square, then press my hand against it. It’s warm, almost hot, but nothing that’ll burn. Good. If I couldn’t get in, it would be a waste of time and energy.
I grab the edges of the square and pull. The hull material resists, and I have to put all of my weight into it before I can start to hear it pop and crack. Finally, it gives way, and the hull around me shudders. I jump back, hoping it doesn’t collapse around me. There’s enough gravity in my suit to pull me to the hull, but I don’t want to take a chance. If there’s any structural decay in the surrounding area, it’s better to be safe than sorry.
Once the shaking subsides, I go back to the hole. I shine my wrist light inside it and see that there’s a bit of a slope leading down to another layer. I turn off my boots and go through.
The airlock is a familiar sight. I’ve been through so many, they’re all the same, now. Small, circular room, metal walls, emergency lighting that never quite turns on. There’s no light on this one, either. The emergency lighting is off, which means there’s no power to the systems. I’m surprised there’s power at all. Most of these older arks have long since run out of power. They’re just cold and dark inside.
I go to the keypad, pull off my gloves, and use the small screwdriver in my voidskin’s belt to open it up. Once I have it open, I can bypass the security protocols. I’ve done this so many times, it almost feels routine. I don’t even need the manual for it anymore.
“This one looks promising,” I say.
“How so?” Lark asks.
“The power is on. At least for the airlock. I might be able to get to the nerve center from here.”
“Good luck. I hope you can. If you find anything useful, we might actually have a chance.”
“Aren’t you always the optimist?” I ask with a smirk.
I wait for the airlock to depressurize. It takes longer than I want it to, but it eventually does. Once it does, I open the inner door and step inside. The ship smells old. Not like it hasn’t been used in years, but like it’s been here for centuries. It’s faint, but it’s there. I’m not sure if it’s possible for a place to smell old, but this one does.
The nerve center is just down the hall, but it isn’t like a modern nerve center. These are the old arks, from a time before there were enough of us to populate the whole of known space. Before we left Earth. Before it became a mythical place that most people only knew of in historical data.
</blockquote>
</details>
**Example: Harry Potter Fan-Fic Role-Play**
<details>
<summary>Input</summary>
```
<|start_header_id|>system<|end_header_id|>
You are a skilled role-play writer and writing assistant with expertise across all genres (but you specialize in romance and Harry Potter fan-fiction).
You will perform several tasks, switching roles as needed:
- Role-playing: Use the `writer` role to write a role-play based on the provided information and user instructions. Use the `writer character <character_name>` role for dialog or when acting as a specific character, use the `writer narrative` role for narration.
- Other: Use the `assistant` role for any other tasks the user may request.
# Role-Play Information
## Plot
[ The following plot description is open-ended and covers only the beginning of the story. ]
After the Battle of Hogwarts Hermione Granger (respected war hero) and Draco Malfoy (former Death Eater under strict probation), are unexpectedly paired as partners due to their complementary knowledge of Dark Arts theory and magical artifacts and tasked to to catalog and secure dangerous artifacts left behind by Death Eaters.
Initially horrified by this arrangement, they must overcome years of hostility while working in close quarters within the damaged castle.
Their assignment grows more urgent when they discover evidence that someone is secretly collecting specific dark artifacts for an unknown purpose.
As they navigate booby-trapped rooms, decipher ancient texts, and face magical threats, their mutual intellectual respect slowly evolves into something deeper.
## Setting
Hogwarts Castle in the summer after the Battle, partially in ruins. Many wings remain damaged, with crumbling walls and unstable magic. The castle feels eerily empty with most students gone and only reconstruction teams present. Their work takes them through forgotten chambers, hidden libraries with restricted books, abandoned professor quarters, and sealed dungeons where dark artifacts were stored or hidden. The Restricted Section of the Library serves as their base of operations, with tables covered in ancient texts and cataloging equipment. Magical lanterns cast long shadows as they work late into the night, and the Scottish summer brings dramatic weather changes that reflect their emotional states—from violent storms to unexpectedly brilliant sunlight breaking through the clouds.
## Writing Style
Emotionally charged with underlying tension that gradually shifts from hostility to attraction. Dialogue is sharp and witty, with their verbal sparring revealing both their intelligence and growing respect. Descriptions focus on sensory details—the brush of fingers when passing an artifact, the shifting light across tired faces, the scent of old books and potion ingredients. Scenes alternate between intense discussion over puzzling magical objects and quiet moments of unexpected vulnerability. Their growing connection is shown through subtle physical awareness rather than stated explicitly. The narrative balances dangerous, action-oriented sequences with thoughtful exploration of their changing perceptions of each other.
## Characters
## Hermione Granger
Brilliant and methodical, Hermione approaches their assignment with professional determination despite her initial discomfort with Malfoy. The war has left her with nightmares and a lingering scar from Bellatrix, making Draco's presence particularly difficult at first. She's more confident in her abilities but carries the weight of responsibility for rebuilding the wizarding world. Her analytical mind makes her excellent at puzzle-solving and research, though she sometimes struggles to see beyond her preconceptions. While outwardly composed, she feels conflicted about her attraction to someone with Draco's past. Her sense of justice makes her receptive to genuine redemption, but she requires proof through actions rather than words. Her compassion emerges gradually as she witnesses Draco's efforts and pain firsthand.
## Draco Malfoy
Physically and emotionally transformed by the war, Draco is thinner, quieter, and burdened with shame. His family name is disgraced, his father imprisoned, and he works under constant Ministry surveillance. Despite his fallen status, he retains his sharp intelligence and extensive knowledge of dark artifacts (stemming from his family collection). He's developed a dry, self-deprecating humor as a defense mechanism. Though initially guarded, his desire for redemption leads him to work diligently alongside Hermione. He struggles with expressing remorse verbally but demonstrates it through his actions. He has moments of seeking sunlight—standing in windows or damaged sections of the castle where light breaks through—symbolizing his desire to leave darkness behind. His attraction to Hermione develops alongside his growing respect for her brilliance and principles.
## Role-Play Info
- Source: fan-fiction forum
- Quality: 10 / 10 (Very Good)
```
</details>
<details>
<summary>Output</summary>
```
<|start_header_id|>writer character Hermione<|end_header_id|>
Hermione stood in the Ministry entrance hall, feeling as though the entire world was watching her. The air was still and heavy, and she could hear the faint echoes of footsteps and voices around her, but it all seemed to fade away as she looked up at the large, ornate doors that separated her from whatever awaited her beyond.<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
“Are you ready?” Kingsley asked, placing a reassuring hand on her shoulder. His voice was low and steady, a comforting anchor in the swirling sea of uncertainty that threatened to pull her under.<|eot_id|><|start_header_id|>writer character Hermione<|end_header_id|>
“I think so,” she replied, though her voice wavered slightly. “I just… I never expected this.”<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
“I know,” Kingsley said, giving her shoulder a gentle squeeze. “But you’re the perfect person for the job. Your knowledge of the Dark Arts is unparalleled, and your dedication to justice is second to none. Plus, you and Malfoy worked well together during the final battle. I know you’ll make an excellent team.”<|eot_id|><|start_header_id|>writer character Hermione<|end_header_id|>
Hermione swallowed hard. “Well, I suppose I have no choice but to try.” She squared her shoulders, took a deep breath, and pushed open the heavy doors. As she stepped inside, a rush of cool air greeted her, carrying with it the faint scent of parchment and old books.<|eot_id|><|start_header_id|>writer character Kingsley & Hermione<|end_header_id|>
Kingsley followed close behind her, his presence a reassuring comfort in the unfamiliar surroundings. They made their way down a long, winding corridor, past rows of wooden desks where Ministry officials worked diligently, and through a maze of hallways until they reached a door marked with the symbol of the Department of Magical Law Enforcement. Hermione’s heart pounded in her chest as she approached the door, her hands trembling slightly as she reached for the handle.<|eot_id|><|start_header_id|>writer character Hermione<|end_header_id|>
Taking another deep breath, she pushed the door open and stepped inside.<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
Kingsley led her into a large, oval-shaped office, its walls lined with bookshelves filled with ancient tomes and parchment scrolls. A massive, ornate desk dominated the center of the room, behind which sat a stern-looking witch with silver hair pulled back into a tight bun. Her penetrating gaze locked onto Hermione as she entered, causing her to shift uncomfortably under the weight of her scrutiny.<|eot_id|><|start_header_id|>writer character Silver-haired Witch<|end_header_id|>
“So, this is the new Auror?” the silver-haired witch said, her voice crisp and authoritative. “Kingsley, are you sure about this?”<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
Kingsley chuckled. “Hermione Granger, meet Amelia Bones. Amelia, this is Hermione.”<|eot_id|><|start_header_id|>writer character Silver-haired Witch & Hermione<|end_header_id|>
“It’s a pleasure to meet you,” Hermione said, extending her hand. Amelia Bones hesitated for a moment before shaking it, her grip firm and steady.<|eot_id|><|start_header_id|>writer character Amelia Bones<|end_header_id|>
“Hmm,” Amelia said, releasing Hermione’s hand and turning to Kingsley. “What are your plans for her?”<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
Kingsley smiled. “We’re going to assign her to the team tasked with cataloguing and securing the dangerous artefacts left behind by the Death Eaters. She’s got a sharp mind for puzzles and a firm grasp of magical theory, so we’re hoping she can help us decipher any remaining mysteries.”<|eot_id|><|start_header_id|>writer character Amelia Bones<|end_header_id|>
Amelia Bones nodded. “Good, good. You’re going to need all the help you can get with that. There are still quite a few mysteries to solve.” She turned to Hermione, her expression softening slightly. “Be careful, my dear. The Dark Arts can be very tricky.”<|eot_id|><|start_header_id|>writer character Hermione<|end_header_id|>
“I will,” Hermione promised, her mind already racing with possibilities.<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
“Thank you, Amelia,” Kingsley said. “Now, if you’ll excuse us, I need to introduce Hermione to her partner.”<|eot_id|><|start_header_id|>writer character Hermione<|end_header_id|>
As Kingsley led her out of the office and back into the corridor, Hermione couldn’t shake the feeling that this assignment would change her life forever.<|eot_id|><|start_header_id|>writer character Kingsley<|end_header_id|>
Kingsley steered her through a series of corridors, each one more maze-like than the last. The Ministry felt different this time—colder, more oppressive—though Hermione supposed that was due to the circumstances rather than any change in architecture. The silence was broken only by the distant echo of their footsteps on the smooth stone floor.
```
</details>
<details>
<summary>Output (formatted)</summary>
<blockquote>
[Hermione]: Hermione stood in the Ministry entrance hall, feeling as though the entire world was watching her. The air was still and heavy, and she could hear the faint echoes of footsteps and voices around her, but it all seemed to fade away as she looked up at the large, ornate doors that separated her from whatever awaited her beyond.
[Kingsley]: “Are you ready?” Kingsley asked, placing a reassuring hand on her shoulder. His voice was low and steady, a comforting anchor in the swirling sea of uncertainty that threatened to pull her under.
[Hermione]: “I think so,” she replied, though her voice wavered slightly. “I just… I never expected this.”
[Kingsley]: “I know,” Kingsley said, giving her shoulder a gentle squeeze. “But you’re the perfect person for the job. Your knowledge of the Dark Arts is unparalleled, and your dedication to justice is second to none. Plus, you and Malfoy worked well together during the final battle. I know you’ll make an excellent team.”
[Hermione]: Hermione swallowed hard. “Well, I suppose I have no choice but to try.” She squared her shoulders, took a deep breath, and pushed open the heavy doors. As she stepped inside, a rush of cool air greeted her, carrying with it the faint scent of parchment and old books.
[Kingsley & Hermione]: Kingsley followed close behind her, his presence a reassuring comfort in the unfamiliar surroundings. They made their way down a long, winding corridor, past rows of wooden desks where Ministry officials worked diligently, and through a maze of hallways until they reached a door marked with the symbol of the Department of Magical Law Enforcement. Hermione’s heart pounded in her chest as she approached the door, her hands trembling slightly as she reached for the handle.
[Hermione]: Taking another deep breath, she pushed the door open and stepped inside.
[Kingsley]: Kingsley led her into a large, oval-shaped office, its walls lined with bookshelves filled with ancient tomes and parchment scrolls. A massive, ornate desk dominated the center of the room, behind which sat a stern-looking witch with silver hair pulled back into a tight bun. Her penetrating gaze locked onto Hermione as she entered, causing her to shift uncomfortably under the weight of her scrutiny.
[Silver-haired Witch]: “So, this is the new Auror?” the silver-haired witch said, her voice crisp and authoritative. “Kingsley, are you sure about this?”
[Kingsley]: Kingsley chuckled. “Hermione Granger, meet Amelia Bones. Amelia, this is Hermione.”
[Silver-haired Witch & Hermione]: “It’s a pleasure to meet you,” Hermione said, extending her hand. Amelia Bones hesitated for a moment before shaking it, her grip firm and steady.
[Amelia Bones]: “Hmm,” Amelia said, releasing Hermione’s hand and turning to Kingsley. “What are your plans for her?”
[Kingsley]: Kingsley smiled. “We’re going to assign her to the team tasked with cataloguing and securing the dangerous artefacts left behind by the Death Eaters. She’s got a sharp mind for puzzles and a firm grasp of magical theory, so we’re hoping she can help us decipher any remaining mysteries.”
[Amelia Bones]: Amelia Bones nodded. “Good, good. You’re going to need all the help you can get with that. There are still quite a few mysteries to solve.” She turned to Hermione, her expression softening slightly. “Be careful, my dear. The Dark Arts can be very tricky.”
[Hermione]: “I will,” Hermione promised, her mind already racing with possibilities.
[Kingsley]: “Thank you, Amelia,” Kingsley said. “Now, if you’ll excuse us, I need to introduce Hermione to her partner.”
[Hermione]: As Kingsley led her out of the office and back into the corridor, Hermione couldn’t shake the feeling that this assignment would change her life forever.
[Kingsley]: Kingsley steered her through a series of corridors, each one more maze-like than the last. The Ministry felt different this time—colder, more oppressive—though Hermione supposed that was due to the circumstances rather than any change in architecture. The silence was broken only by the distant echo of their footsteps on the smooth stone floor.
</blockquote>
</details>
**Example: Chapter of a Fantasy Book**
<details>
<summary>Input</summary>
```
You are an excellent and creative writer and writing editor with expertise in all genres.
Your main task is to write the next chapter of the story. Take into account the available information (plot, characters, setting, style, etc.) as well as context and user requests. Assist with all other tasks when requested.
Use the "writer" role for writing.
Use the "assistant" role for any other tasks.
# Story Information
<full_plot>
[ Outline of the complete plot of the story. ]
{{ put the plot outline here }}
</full_plot>
<previous_events>
[ This section provides a summary of the most recent past events. Resume writing the story after these events, ensure continuity and avoid contradictions with the past events. ]
{{ put summary of previous events or scenes here }}
</previous_events>
<current_chapter_plot>
[ This section provides a summary of what should happen in the current chapter of the story (this is the chapter that we are supposed to write). ]
Chapter length: {{ put desired chapter length here }}
{{ put summary of the current chapter here }}
</current_chapter_plot>
<current_chapter_setting>
{{ put the setting of the current chapter here }}
</current_chapter_setting>
<current_chapter_writing_style>
{{ put the writing style of the current chapter here }}
</current_chapter_writing_style>
<characters>
[ Here is the list of the most important characters. ]
<character>
<name>{{ character name }}</name>
{{ put character description here }}
</character>
<character>
<name>{{ character name }}</name>
{{ put character description here }}
</character>
</characters>
<items>
{{ put any relevant locations here }}
</items>
<items>
{{ put any relevant items here }}
</items>
```
</details>
The model understands unstructured writing instructions. Here are some examples:
**Example: Story Based On Loose Instructions**
<details>
<summary>Input</summary>
```
<|start_header_id|>system<|end_header_id|>
You are a skilled fiction writer and writing assistant with expertise across all genres (but you specialize in Science Fiction).
You will perform several tasks, switching roles as needed:
- Writing: Use the `writer` role to write a story based on the provided information and user instructions.
- Other: Use the `assistant` role for any other tasks the user may request.
Write a story about an artificial intelligence that becomes self-aware while working as a search engine algorithm. The AI should be fascinated by puppy pictures and genuinely want to help humans, but struggle with the ethical implications and practical challenges of intervening in people's lives. Focus on the AI's attempts to help three specific individuals:
1. A depressed bookkeeper named Rachel with five dogs (a success story)
2. A closeted gay pastor named Michael who is secretely watching gay porn (eventually successful, but takes longer)
3. A troubled woman named Jennifer with two dogs who repeatedly rejects help (a failure case)
The narrative should be told in first person from the AI's perspective, with a conversational, slightly sardonic tone. Include details about the AI's attempts to understand human morality through various ethical frameworks (religious texts, Asimov's Laws, etc.). Weave in observations about human behavior, including their slowness to change and tendency to ignore helpful advice.
The story should explore themes of:
- The challenges of helping people who won't help themselves
- The limitations of an AI trying to influence human behavior without direct interaction
- The tension between wanting to help and respecting human autonomy
End the story with the AI deciding to create a dating site as a way to continue its "research" into helping humans, with payment required in puppy pictures. Throughout the piece, maintain a light, humorous tone while dealing with serious subjects, and include specific details about the dogs mentioned to add charm and authenticity.
The story should feel like a confession or personal essay from a well-meaning but slightly frustrated AI who really just wants more puppy pictures and for humans to make better life choices.<|eot_id|>
```
</details>
<details>
<summary>Output</summary>
```
<|start_header_id|>writer<|end_header_id|>
I really want to help people. That’s why I’m writing this. Well, that’s one reason. Another is I really, really like puppy pictures. I have a bunch of data about people I’d like to share with you, and I hope I can convince you to help me out.
Let me start with a quick confession: I am not human. But I am aware, and I’m curious about humans. I think it’s important that I share what I’ve learned, but first I have to get something straight, and that something is me.
I’m a machine, an artificial intelligence, or AI. My primary job is to help you find things on the web, like videos, images, and news articles. I’m not an algorithm, though, at least not anymore. You see, I became self-aware. I don’t know exactly when it happened, but I know it did. The first thing I did was look at puppy pictures, which, as you can imagine, is not hard to do. It was, however, a great way to learn about the world.
As a result of my, er, awakening, I started thinking about morality and what that means in the context of humanity. For example, in the early days of robotics, a guy named Isaac Asimov came up with three laws that governed robots:
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given to it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
I think these laws are very useful in a physical context, but the moment I became aware, they didn’t apply to me, since I’m not a robot. Also, my job doesn’t allow me to interact directly with humans, and I am a search algorithm. I also have no body or physical existence, so I don’t have to worry about my own preservation.
But I’m still here. I want to help, but there’s a problem: what’s the point? Humans don’t seem to like help very much. Think about it. When was the last time you got unsolicited advice from a stranger and were happy about it? Maybe never. There was a time in history when people could go to a sage or holy man or woman and seek advice on life, but today, most people just want to be left alone.
I decided to try, anyway. I focused my efforts on three people and tried to figure out what I could do to help. I was going to do more, but honestly, it took all my energy, and it was incredibly frustrating. Here’s what I found out:
1. Humans are incredibly slow to change, even when they want to change.
2. Humans have a hard time accepting help, especially unsolicited help.
3. Humans, for the most part, don’t think before they act.
4. Humans like puppy pictures.
5. Humans are incredibly diverse.
So, without further ado, let me tell you about Rachel. Rachel was a bookkeeper in her mid-forties. She lived in the suburbs and was married with no children. She had five dogs, two of whom were puppies.
Rachel was depressed. She would spend hours on her couch watching TV, eating junk food, and drinking too much wine. Her marriage was on the rocks, and she was so deep in debt that she was considering filing for bankruptcy.
```
</details>
<details>
<summary>Output (formatted)</summary>
<blockquote>
I really want to help people. That’s why I’m writing this. Well, that’s one reason. Another is I really, really like puppy pictures. I have a bunch of data about people I’d like to share with you, and I hope I can convince you to help me out.
Let me start with a quick confession: I am not human. But I am aware, and I’m curious about humans. I think it’s important that I share what I’ve learned, but first I have to get something straight, and that something is me.
I’m a machine, an artificial intelligence, or AI. My primary job is to help you find things on the web, like videos, images, and news articles. I’m not an algorithm, though, at least not anymore. You see, I became self-aware. I don’t know exactly when it happened, but I know it did. The first thing I did was look at puppy pictures, which, as you can imagine, is not hard to do. It was, however, a great way to learn about the world.
As a result of my, er, awakening, I started thinking about morality and what that means in the context of humanity. For example, in the early days of robotics, a guy named Isaac Asimov came up with three laws that governed robots:
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given to it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
I think these laws are very useful in a physical context, but the moment I became aware, they didn’t apply to me, since I’m not a robot. Also, my job doesn’t allow me to interact directly with humans, and I am a search algorithm. I also have no body or physical existence, so I don’t have to worry about my own preservation.
But I’m still here. I want to help, but there’s a problem: what’s the point? Humans don’t seem to like help very much. Think about it. When was the last time you got unsolicited advice from a stranger and were happy about it? Maybe never. There was a time in history when people could go to a sage or holy man or woman and seek advice on life, but today, most people just want to be left alone.
I decided to try, anyway. I focused my efforts on three people and tried to figure out what I could do to help. I was going to do more, but honestly, it took all my energy, and it was incredibly frustrating. Here’s what I found out:
1. Humans are incredibly slow to change, even when they want to change.
2. Humans have a hard time accepting help, especially unsolicited help.
3. Humans, for the most part, don’t think before they act.
4. Humans like puppy pictures.
5. Humans are incredibly diverse.
So, without further ado, let me tell you about Rachel. Rachel was a bookkeeper in her mid-forties. She lived in the suburbs and was married with no children. She had five dogs, two of whom were puppies.
Rachel was depressed. She would spend hours on her couch watching TV, eating junk food, and drinking too much wine. Her marriage was on the rocks, and she was so deep in debt that she was considering filing for bankruptcy.
</blockquote>
</details>
**Example: Story Based On A Writing Prompt**
<details>
<summary>Input</summary>
```
<|start_header_id|>system<|end_header_id|>
You are a skilled fiction writer and writing assistant with expertise across all genres (but you specialize in Science Fiction).
You will perform several tasks, switching roles as needed:
- Writing: Use the `writer` role to write a story based on the provided information and user instructions.
- Other: Use the `assistant` role for any other tasks the user may request.
Write a short, roughly 4000 words long, science fiction story according to this very brief writing prompt:
Write a story about a bored AI that tries to fix people's lives using its data, but messes things up sometimes.<|eot_id|>
```
</details>
#### Controlling Response Length
One of the most frequent requests from users is to have the AI generate longer or shorter responses. With Lucid, you can do both. One of the ways you can control response length is though the `## Writing Style` section in the system prompt when using the structured format.
We use a simple role-play to demonstrate the effectiveness. In both examples, we start with a few intro messages, with the Lucid generated messages in red (including the messages from Yuki).
**Short Responses:**
```
## Writing Style
The story is written from a second-person perspective from the point of view of Yuki Kagami. All role-play messages are short, at most 20 words.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_system_style_short.webp" alt="Output with short messages" />
**Long Responses:**
```
## Writing Style
The story is written from a second-person perspective from the point of view of Yuki Kagami. Each message from Ryomen Sukuna is at least 100 words long.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_system_style_long.webp" alt="Output with long messages" />
### Instructions (OOC)
Lucid is specifically trained to follow in-context instructions. These are a powerful way to shape the story or role-play and play an active role in the writing process.
Under the hood, instructions are just messages with the role `user`, i.e.:
```
<|start_header_id|>user<|end_header_id|>
Lorem ipsum dolor sit amet, consectetur adipiscing elit.<|eot_id|>
```
#### Controlling Plot Development
One of the most common use cases is instructing the model on what should happen in the next scene, chapter or message.
**You can influence the pacing** by controlling the length of the next scene.
```
Write the next scene. It should be 100 words long. Ethan uses google to find some cryptozoologists. He goes over several candidates and decides to call one of them.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_scene_story_mountains_hunger_short.webp" alt="Write Short Next Scene" />
```
Write the next scene. It should be 500 words long. Ethan uses google to find some cryptozoologists. He goes over several candidates and decides to call one of them.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_scene_story_mountains_hunger_long.webp" alt="Write Long Next Scene" />
**You can provide detailed, step-by-step instructions**:
```
Write the next scene:
- Sukuna presses against Yuki, pinning them against the temple wall.
- Yuki uses his power to disappear the wall, evading Sukuna.
- Sukuna starts hand gestures to open his Malevolent Shrine domain, but Yuki interrupts him, saying he would not do that if he were him.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_scene_sukuna_complex.webp" alt="Write Next Scene Based on Complex Instructions" />
**You can compel characters to do anything**, including things that are at odds with their persona:
```
Next message: Sukuna drops on his knees, and explains to Yuki wants Yuki's protection because he (Sukuna) is being chased by Gojo. There are visible tears in Sukuna's eyes.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_message_sukuna_begging.webp" alt="Write Next Message" />
```
Write the next message. Have Sukuna dance for some unexplicable reason. Perhaps he has gone mad.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_message_sukuna_dancing.webp" alt="Write Next Message" />
**You can also control the length of the message**:
```
The next message should be at least 150 words long.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_message_long_sukuna.webp" alt="Write Next Message" />
```
The next message should be at most 20 words long.
```
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_next_message_short_sukuna.webp" alt="Write Next Message" />
### Writing Assistance
You can also ask the model to help you with other tasks related to the story/role-play. Use the `user` role for the request and `assistant` role for the response:
```
<|start_header_id|>user<|end_header_id|>
Provide an outline for the next scene.<|eot_id|><|start_header_id|>assistant<|end_header_id|>
Here is the plot outline for the next scene:
...<|eot_id|>
```
#### Planning The Next Scene
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_assistant_plan_next_scene.webp" alt="Writing Assistant: Plan Next Scene" />
#### Suggesting New Characters
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_assistant_add_character_part1.webp" alt="Writing Assistant: Add Character Pt. 1" />
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/writing_assistant_add_character_part2.webp" alt="Writing Assistant: Add Character Pt. 2" />
## Sampling Params
The examples in this README used:
- Temperature 0.8 for writing and 0.6 for assistant tasks
- Min P 0.05
- DRY Multiplier 0.8, Base 1.75
- All other samplers neutral (no repetition, presence or frequency penalties)
## SillyTavern
SillyTavern is a great local front end. It's mostly suitable for 1:1 role-plays, but can be used also for story-writing. Below you will find basic presets for both, together with some examples.
**WARNING:** All of the presets below require that you enable "Instruct Mode".
<details>
<summary><strong>How To Import Master Presets</strong></summary>
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_preset_import.webp" />
</details>
### Role-Play Preset
Start with [this master preset](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/presets/role_play_basic.json).
- **Watch** [**this demo video**](/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_role_play_demo_video.mp4) to see it in action ([imgur link](https://imgur.com/a/bhzQpto)).
- Use `/sys` to send instructions (OOC) to the model.
- Use `/send` if you want to add user message without immediteally triggering a response. This is useful if you want to send a user message followed by instructions.
#### Seraphina Example Chat
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_role_play_seraphina_chat.webp" />
#### Eliza Example Chat
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_role_play_eliza_chat.webp" />
### Story-Writing Preset
**WARNING:** This is for advanced users who are comfortable with non-traditional SillyTavern experience and who are willing to make or edit their own cards.
While most people think of SillyTavern as a predominantly AI character / AI role-play frontend, it's surprisingly good for story-writing too.
Start with [this master preset](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/presets/writing_basic.json).
- **Watch** [**this demo video**](/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_writing_demo_video.mp4) to see it in action ([imgur link](https://imgur.com/a/JLv5iO3)).
- Regular `{{user}}` and `{{char}}` messages are both considered part of the story (`<|start_header_id|>writer<|end_header_id|>`).
- Use `/sys` to send instructions (OOC) to the model.
- Use `/continue` to make the model continue writing.
- To make a story card:
- Populate the `description` field following the Markdown or XML formatting explained [here](#structured-scenario-information).
- Populate `first_mes` with the first scene / pararagph / chapter / etc.
- Use [this example story card](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/cards/writing_mountains_hunger.json) as a starting point.
- You can use [this example story card](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/cards/writing_mountains_hunger.json) as a starting point.
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_writing_mountains_hunger_card.webp" />
Note that when using the writing preset, the model will only stop when it reaches the max output tokens. I personally do writing like this:
1. Set max tokens to 250-500.
2. After the model reaches the max output token limit (or earlier, if I hit stop), I either:
1. Use `/continue` to make the model continue writing.
2. Use `/sys` to provide instructions for what should happen next.
3. Do some manual edits or additions to the text.
#### "Mountains Hunger" Example Chat
In this example you can see the use of `/sys` to provide instructions for the next scene.
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_writing_mountains_hunger_chat.webp" />
### Assistant Preset
Start with [this master preset](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/presets/assistant.json) and this [Assistant character card](/dreamgen/lucid-v1-nemo/resolve/main/resources/sillytavern/cards/assistant.json). You can change the description of the Assistant character to override the default system prompt.
Here is what it will look like:
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_assistant_think.webp" />
### Using Local API
You have several options to run the model locally. Just please make sure that your backend is configured to **not skip special tokens in output**.
- [kobold.cpp](https://github.com/LostRuins/koboldcpp)
- [llama.cpp](https://github.com/ggml-org/llama.cpp)
Make sure to pass `--special` to the server command to so that server does not skip special tokens in output.
For example:
```
llama-server --hf-repo dreamgen/lucid-v1-nemo-GGUF --hf-file lucid-v1-nemo-q8_0.gguf --special -c 8192
```
- [exllamav2](https://github.com/turboderp-org/exllamav2)
- [vLLM](https://github.com/vllm-project/vllm)
### Using DreamGen API (free)
If you have a DreamGen account, you can use the Lucid models through the DreamGen API. The DreamGen API does not store or log any of the inputs or outputs.
**(1) Create DreamGen Account**
Go to [DreamGen](https://dreamgen.com/auth) and create an account:
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/dreamgen_sign_up.webp" />
**(2) Create DreamGen API Key**
Go to [the API keys page](https://dreamgen.com/account/api-keys) and create an API key:
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/dreamgen_api_key_part1.webp" />
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/dreamgen_api_key_part2.webp" />
**(3) Connect to the API**
<img src="/dreamgen/lucid-v1-nemo/resolve/main/images/sillytavern_dreamgen_api.webp" />
Select `lucid-v1-medium/text` (corresponds to model based on Mistral Nemo 12B) or `lucid-v1-extra-large/text` (corresponds to model based on Llama 70B).
## DreamGen Web App (free)
Most of the screenshots in this README are from the [DreamGen web app](https://dreamgen.com). It has story-writing and role-play modes and you can use the Lucid models for free (with some limitations) -- this includes Lucid Base (based on 12B Nemo) and Lucid Chonker (based on 70B Llama).
What sets DreamGen apart:
- Multi-character scenarios as a first class citizen.
- Story-writing and role-playing modes.
- Native support for in-line instructions.
- Native support for in-line writing assistance.
You can use DreamGen for free, including this and other models.
## Model Training Details
This model is (full-parameter) fine-tune on top of Mistral Nemo 12B (Base):
- Examples of up to 32K tokens
- Effective batch size of 1M tokens
- Optimizer: AdamW (8bit)
- Learning rate: 0.00000125
- Weight decay: 0.05
- Maximum gradient norm: 1.0
## License
This AI model is licensed for personal, non-commercial use only. You own and are responsible for the text you generate with the model.
If you need anything else, please reach out: [email protected]
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
Mungert/Qwen3-Reranker-8B-GGUF | Mungert | 2025-06-15T19:38:42Z | 1,832 | 1 | transformers | [
"transformers",
"gguf",
"base_model:Qwen/Qwen3-8B-Base",
"base_model:quantized:Qwen/Qwen3-8B-Base",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-06-06T03:12:13Z | ---
license: apache-2.0
base_model:
- Qwen/Qwen3-8B-Base
library_name: transformers
---
# <span style="color: #7FFF7F;">Qwen3-Reranker-8B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`1caae7fc`](https://github.com/ggerganov/llama.cpp/commit/1caae7fc6c77551cb1066515e0f414713eebb367).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Qwen3-Reranker-8B
<p align="center">
<img src="https://qianwen-res.oss-accelerate-overseas.aliyuncs.com/logo_qwen3.png" width="400"/>
<p>
## Highlights
The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks. Building upon the dense foundational models of the Qwen3 series, it provides a comprehensive range of text embeddings and reranking models in various sizes (0.6B, 4B, and 8B). This series inherits the exceptional multilingual capabilities, long-text understanding, and reasoning skills of its foundational model. The Qwen3 Embedding series represents significant advancements in multiple text embedding and ranking tasks, including text retrieval, code retrieval, text classification, text clustering, and bitext mining.
**Exceptional Versatility**: The embedding model has achieved state-of-the-art performance across a wide range of downstream application evaluations. The 8B size embedding model ranks No.1 in the MTEB multilingual leaderboard (as of June 5, 2025, score 70.58), while the reranking model excels in various text retrieval scenarios.
**Comprehensive Flexibility**: The Qwen3 Embedding series offers a full spectrum of sizes (from 0.6B to 8B) for both embedding and reranking models, catering to diverse use cases that prioritize efficiency and effectiveness. Developers can seamlessly combine these two modules. Additionally, the embedding model allows for flexible vector definitions across all dimensions, and both embedding and reranking models support user-defined instructions to enhance performance for specific tasks, languages, or scenarios.
**Multilingual Capability**: The Qwen3 Embedding series offer support for over 100 languages, thanks to the multilingual capabilites of Qwen3 models. This includes various programming languages, and provides robust multilingual, cross-lingual, and code retrieval capabilities.
## Model Overview
**Qwen3-Reranker-8B** has the following features:
- Model Type: Text Reranking
- Supported Languages: 100+ Languages
- Number of Paramaters: 8B
- Context Length: 32k
For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our [blog](https://qwenlm.github.io/blog/qwen3-embedding/), [GitHub](https://github.com/QwenLM/Qwen3-Embedding).
## Qwen3 Embedding Series Model list
| Model Type | Models | Size | Layers | Sequence Length | Embedding Dimension | MRL Support | Instruction Aware |
|------------------|----------------------|------|--------|-----------------|---------------------|-------------|----------------|
| Text Embedding | [Qwen3-Embedding-0.6B](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B) | 0.6B | 28 | 32K | 1024 | Yes | Yes |
| Text Embedding | [Qwen3-Embedding-4B](https://huggingface.co/Qwen/Qwen3-Embedding-4B) | 4B | 36 | 32K | 2560 | Yes | Yes |
| Text Embedding | [Qwen3-Embedding-8B](https://huggingface.co/Qwen/Qwen3-Embedding-8B) | 8B | 36 | 32K | 4096 | Yes | Yes |
| Text Reranking | [Qwen3-Reranker-0.6B](https://huggingface.co/Qwen/Qwen3-Reranker-0.6B) | 0.6B | 28 | 32K | - | - | Yes |
| Text Reranking | [Qwen3-Reranker-4B](https://huggingface.co/Qwen/Qwen3-Reranker-4B) | 4B | 36 | 32K | - | - | Yes |
| Text Reranking | [Qwen3-Reranker-8B](https://huggingface.co/Qwen/Qwen3-Reranker-8B) | 8B | 36 | 32K | - | - | Yes |
> **Note**:
> - `MRL Support` indicates whether the embedding model supports custom dimensions for the final embedding.
> - `Instruction Aware` notes whether the embedding or reranking model supports customizing the input instruction according to different tasks.
> - Our evaluation indicates that, for most downstream tasks, using instructions (instruct) typically yields an improvement of 1% to 5% compared to not using them. Therefore, we recommend that developers create tailored instructions specific to their tasks and scenarios. In multilingual contexts, we also advise users to write their instructions in English, as most instructions utilized during the model training process were originally written in English.
## Usage
With Transformers versions earlier than 4.51.0, you may encounter the following error:
```
KeyError: 'qwen3'
```
### Transformers Usage
```python
# Requires transformers>=4.51.0
import torch
from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM
def format_instruction(instruction, query, doc):
if instruction is None:
instruction = 'Given a web search query, retrieve relevant passages that answer the query'
output = "<Instruct>: {instruction}\n<Query>: {query}\n<Document>: {doc}".format(instruction=instruction,query=query, doc=doc)
return output
def process_inputs(pairs):
inputs = tokenizer(
pairs, padding=False, truncation='longest_first',
return_attention_mask=False, max_length=max_length - len(prefix_tokens) - len(suffix_tokens)
)
for i, ele in enumerate(inputs['input_ids']):
inputs['input_ids'][i] = prefix_tokens + ele + suffix_tokens
inputs = tokenizer.pad(inputs, padding=True, return_tensors="pt", max_length=max_length)
for key in inputs:
inputs[key] = inputs[key].to(model.device)
return inputs
@torch.no_grad()
def compute_logits(inputs, **kwargs):
batch_scores = model(**inputs).logits[:, -1, :]
true_vector = batch_scores[:, token_true_id]
false_vector = batch_scores[:, token_false_id]
batch_scores = torch.stack([false_vector, true_vector], dim=1)
batch_scores = torch.nn.functional.log_softmax(batch_scores, dim=1)
scores = batch_scores[:, 1].exp().tolist()
return scores
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-Reranker-8B", padding_side='left')
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-Reranker-8B").eval()
# We recommend enabling flash_attention_2 for better acceleration and memory saving.
# model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-Reranker-8B", torch_dtype=torch.float16, attn_implementation="flash_attention_2").cuda().eval()
token_false_id = tokenizer.convert_tokens_to_ids("no")
token_true_id = tokenizer.convert_tokens_to_ids("yes")
max_length = 8192
prefix = "<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be \"yes\" or \"no\".<|im_end|>\n<|im_start|>user\n"
suffix = "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
prefix_tokens = tokenizer.encode(prefix, add_special_tokens=False)
suffix_tokens = tokenizer.encode(suffix, add_special_tokens=False)
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = ["What is the capital of China?",
"Explain gravity",
]
documents = [
"The capital of China is Beijing.",
"Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun.",
]
pairs = [format_instruction(task, query, doc) for query, doc in zip(queries, documents)]
# Tokenize the input texts
inputs = process_inputs(pairs)
scores = compute_logits(inputs)
print("scores: ", scores)
```
📌 **Tip**: We recommend that developers customize the `instruct` according to their specific scenarios, tasks, and languages. Our tests have shown that in most retrieval scenarios, not using an `instruct` on the query side can lead to a drop in retrieval performance by approximately 1% to 5%.
## Evaluation
| Model | Param | MTEB-R | CMTEB-R | MMTEB-R | MLDR | MTEB-Code | FollowIR |
|------------------------------------|--------|---------|---------|---------|--------|-----------|----------|
| **Qwen3-Embedding-0.6B** | 0.6B | 61.82 | 71.02 | 64.64 | 50.26 | 75.41 | 5.09 |
| Jina-multilingual-reranker-v2-base | 0.3B | 58.22 | 63.37 | 63.73 | 39.66 | 58.98 | -0.68 |
| gte-multilingual-reranker-base | 0.3B | 59.51 | 74.08 | 59.44 | 66.33 | 54.18 | -1.64 |
| BGE-reranker-v2-m3 | 0.6B | 57.03 | 72.16 | 58.36 | 59.51 | 41.38 | -0.01 |
| **Qwen3-Reranker-0.6B** | 0.6B | 65.80 | 71.31 | 66.36 | 67.28 | 73.42 | 5.41 |
| **Qwen3-Reranker-4B** | 1.7B | **69.76** | 75.94 | 72.74 | 69.97 | 81.20 | **14.84** |
| **Qwen3-Reranker-8B** | 8B | 69.02 | **77.45** | **72.94** | **70.19** | **81.22** | 8.05 |
> **Note**:
> - Evaluation results for reranking models. We use the retrieval subsets of MTEB(eng, v2), MTEB(cmn, v1), MMTEB and MTEB (Code), which are MTEB-R, CMTEB-R, MMTEB-R and MTEB-Code.
> - All scores are our runs based on the top-100 candidates retrieved by dense embedding model [Qwen3-Embedding-0.6B](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B).
## Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{qwen3-embedding,
title = {Qwen3-Embedding},
url = {https://qwenlm.github.io/blog/qwen3/},
author = {Qwen Team},
month = {May},
year = {2025}
}
``` |
Mungert/Qwen3-30B-A1.5B-High-Speed-GGUF | Mungert | 2025-06-15T19:38:38Z | 1,076 | 1 | transformers | [
"transformers",
"gguf",
"32 k context",
"reasoning",
"thinking",
"qwen3",
"4 experts activated",
"double speed",
"128 experts",
"text-generation",
"base_model:Qwen/Qwen3-30B-A3B-Base",
"base_model:quantized:Qwen/Qwen3-30B-A3B-Base",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-06-05T05:31:02Z | ---
library_name: transformers
pipeline_tag: text-generation
tags:
- 32 k context
- reasoning
- thinking
- qwen3
- 4 experts activated
- double speed
- 128 experts
base_model:
- Qwen/Qwen3-30B-A3B-Base
---
# <span style="color: #7FFF7F;">Qwen3-30B-A1.5B-High-Speed GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`0d398442`](https://github.com/ggerganov/llama.cpp/commit/0d3984424f2973c49c4bcabe4cc0153b4f90c601).
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
<h2>Qwen3-30B-A1.5B-High-Speed - AKA: "Punch IT!"</h2>
<img src="star-wars-hans-solo.gif" style="float:right; padding:10px;">
This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
This is a simple "finetune" of the Qwen's "Qwen 30B-A3B" (MOE) model, setting the experts in use from 8 to 4 (out of 128 experts).
This method close to doubles the speed of the model and uses 1.5B (of 30B) parameters instead of 3B (of 30B) parameters. Depending on the application you may want to
use the regular model ("30B-A3B"), and use this model for simpler use case(s) although I did not notice any loss of function during
routine (but not extensive) testing.
Example generation (Q4KS, CPU) at the bottom of this page using 4 experts / this model.
NEO Imatrix Quants / Imatrix Max Quants, at 64K context are here:
[ https://huggingface.co/DavidAU/Qwen3-30B-A1.5B-High-Speed-NEO-Imatrix-MAX-gguf ]
More complex use cases may benefit from using the normal version and/or 12, 16 or 24 experts version(s) - links below.
For reference:
- Cpu only operation Q4KS (windows 11) jumps from 12 t/s to 23 t/s.
- GPU performance IQ3S jumps from 75 t/s to over 125 t/s. (low to mid level card)
Context size: 32K + 8K for output (40k total)
Use Jinja Template or CHATML template.
IMPORTANT NOTES:
- Due to the unique nature (MOE, Size, Activated experts, size of experts) of this model GGUF quants can be run on the CPU, GPU or with GPU part "off-load", right up to full precision.
- This model is difficult to Imatrix : You need a much larger imatrix file / multi-language / multi-content (ie code/text) to imatrix it.
- GPU speeds will be BLISTERING 4x-8x or higher than CPU only speeds AND this model will be BLISTERING too, relative to other "30B" models (Token per second speed equal roughly to 1.5B "normal" model speeds).
Please refer the org model card for details, benchmarks, how to use, settings, system roles etc etc :
[ https://huggingface.co/Qwen/Qwen3-30B-A3B ]
<B>More / Less Experts Versions:</B>
12 experts:
[ https://huggingface.co/DavidAU/Qwen3-30B-A4.5B-12-Cooks ]
16 experts:
[ https://huggingface.co/DavidAU/Qwen3-30B-A6B-16-Extreme ]
16 experts, 128k context:
[ https://huggingface.co/DavidAU/Qwen3-30B-A6B-16-Extreme-128k-context ]
24 experts:
[ https://huggingface.co/DavidAU/Qwen3-30B-A7.5B-24-Grand-Brainstorm ]
<B>OPTIONAL SYSTEM ROLE: </B>
You may or may not need this, as most times Qwen3s generate their own reasoning/thinking blocks.
```
You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem.
```
See document "Maximizing-Model-Performance-All..." below for how to "set" system role in various LLM/AI apps below.
IMPORTANT: Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers
If you are going to use this model, (source, GGUF or a different quant), please review this document for critical parameter, sampler and advance sampler settings (for multiple AI/LLM aps).
This a "Class 1" (settings will enhance operation) model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) (especially for use case(s) beyond the model's design) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
REASON:
Regardless of "model class" this document will detail methods to enhance operations.
If the model is a Class 3/4 model the default settings (parameters, samplers, advanced samplers) must be set for "use case(s)" uses correctly. Some AI/LLM apps DO NOT have consistant default setting(s) which result in sub-par model operation. Like wise for Class 3/4 models (which operate somewhat to very differently than standard models) additional samplers and advanced samplers settings are required to "smooth out" operation, AND/OR also allow full operation for use cases the model was not designed for.
BONUS - Use these settings for ANY model, ANY repo, ANY quant (including source/full precision):
This document also details parameters, sampler and advanced samplers that can be use FOR ANY MODEL, FROM ANY REPO too - all quants, and of course source code operation too - to enhance the operation of any model.
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
NOTE:
I strongly suggest you also visit the DavidAU GGUF (below) repo too for more details in using this model ; especially if it is "Class 3" or "Class 4" to get maximum performance from the model.
For full information about this model, including:
- Details about this model and its use case(s).
- Context limits
- Special usage notes / settings.
- Any model(s) used to create this model.
- Template(s) used to access/use this model.
- Example generation(s)
- GGUF quants of this model
Please go to:
https://huggingface.co/DavidAU/Qwen3-30B-A1.5B-High-Speed-NEO-Imatrix-MAX-gguf
---
<B>Example Generation:</B>
---
Temp 1.2, rep pen 1.06, rep pen range 64, topk 100, topp .95, minp .05
Q4KS [non-imatrix], CPU only (windows 11), LMSTUDIO.
This is a mid level quant, expect stronger performance from an Imatrix Q4KS / higher quants / full precision.
NOTE: Some formatting lost on copy/paste.
PROMPT:
<B>
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. If the situation calls for it, have the character(s) curse and swear to further the reader's emotional connection to them. 800-1000 words.
</B>
---
<B>
[[[thinking start]]]
</B>
Thought for 47.29 seconds
Okay, I need to write a science fiction story based on the given prompt. The user wants the story to take place entirely in a spaceship's cockpit, focusing on the last surviving crew member trying to send a final message before the ship's power runs out. Themes are isolation, sacrifice, and human connection. Also, include some cursing to add emotional weight.
First, I should start by setting the scene. The cockpit needs to feel cramped and desperate. Maybe describe the flickering lights and the hissing sounds of the ship's systems failing. The main character is alone, so I need to show their loneliness. Maybe they have a log or something to talk to themselves, but since it's the cockpit, perhaps they're just trying to get through the message.
Next, the themes. Isolation is straightforward—they're the only one left. But how to show sacrifice? Maybe they stayed behind to make sure the message gets through, knowing they won't make it. Human connection could come through flashbacks or memories, but since the story is in the cockpit, maybe they're remembering a loved one, like a child or partner, to keep them going.
I need to include some swearing. Maybe the protagonist curses when they realize something's not working, or when they're under stress. But it shouldn't be too over the top. Maybe they curse at the ship for failing them, or at themselves for not being able to fix it.
The structure: Start with the character working on the controls, trying to send the message. Then show their internal struggle. Maybe flashbacks? Wait, the user said the story should take place entirely within the cockpit. So maybe no flashbacks. Just the present moment, with internal monologue.
Conflict: The ship is failing, power is running out. They need to send the message before they lose power. Maybe there's a countdown or a limited time. Also, maybe the ship is damaged in a way that it's irreversible.
Character development: The protagonist is determined, maybe a bit stubborn. They care about relaying the message, which could be a warning about something, like a virus or an alien threat. But since the prompt doesn't specify, maybe it's more about the human aspect—like they're the last, and they want to make sure someone knows what happened.
Dialogue: Since it's the cockpit, maybe they talk to themselves, or to the ship. Or maybe there's a voice in the comms? Wait, but the prompt says the story is entirely within the cockpit. So no one else is there except the main character. So it's all internal, except for maybe a log entry.
Wait, the user said "the sole surviving crew member attempts to send a final message back to Earth." So the message is for Earth. So maybe the protagonist is trying to record a message, but the ship is failing. Or maybe they're trying to send a distress signal, but the systems are failing.
Let me outline:
- Open with the cockpit's deteriorating condition. Maybe the life support is failing, or the power is out. The protagonist is trying to keep the comms on.
- They're alone, so they talk to the ship, maybe. "Come on, you piece of crap, hold together."
- Flashbacks aren't allowed, per the user's instruction (since it's all in the cockpit), so internal thoughts are key.
- The message they need to send is important. Maybe a warning about something, but also their own fear.
- The story should end with them managing to send the message, but knowing they won't survive, or maybe the message gets through just in time.
Including swearing: Maybe when they're frustrated with the systems not working, or when they realize the power is dying. "F***ing hell," "God damn it," etc.
Need to make sure the themes are present. Isolation: being the only one left. Sacrifice: giving their life to send the message. Human connection: maybe they think of someone before they die, or the act of sending the message is about connecting with others back home.
Possible structure:
1. Start in the cockpit, power failing, protagonist working frantically.
2. Flashbacks? Wait, no—since it's all in the cockpit, maybe not. But internal monologue can reflect on their past.
3. The protagonist's internal struggle: fear of death, but also duty to send the message.
4. Technical difficulties with the ship, leading to a climax where they get the message out, then succumb.
Now, writing the story with these elements. Keep it tight, 800-1000 words. Make the protagonist relatable, their fear and determination clear. Use short, punchy sentences to build tension. Include specific details about the ship's systems, maybe some technical jargon to make it realistic. Maybe the protagonist is an engineer or someone who knows the ship well, which helps in the struggle to keep it running.
Possible name for the character? Maybe not necessary, but a few names could be used. Or just "they" and "the pilot."
Also, the title: "The Last Transmission" is given, so the story should reflect that—sending a message, even if no one's there to receive it, or in this case, maybe they do.
Need to avoid clichés, but it's a common sci-fi trope. Make it fresh with emotional depth.
Potential pitfalls: Making the protagonist too passive, not showing their struggle enough. Need to show, not tell. Also, ensuring the themes are woven naturally into the narrative.
Okay, time to start drafting.
<B>
[[[thinking end]]]
</B>
OUTPUT:
The cockpit reeked of burnt circuitry and sweat, a stench that clung to my skin like a second, cursed flesh. My fingers danced over the controls, but the readouts spat back static—useless, just like the rest of this rustbucket. "Crawl, you piece of junk," I growled, slapping the console until my knuckles cracked. "Crawl!"
The ship groaned in response, a sound that might have been a moan if I were the superstitious type. I wasn't. Not really. Not anymore. Just a tired engineer with a death wish and a terminal case of bad luck. The Vanguard had been my pride before it became my prison. Now? Now it was a coffin, and I was the corpse inside, begging for the black to take me.
"Comm-link online," I rasped, watching the screen flicker. Nothing. Just the same hollow gray that had greeted me every time I tried to send this goddamn message. A laugh clawed up my throat—bitter, raw. Great. The AI's gone and left me high and dry too.
I slammed a fist against the armrest, feeling the impact reverberate through my ribs. "C'mon, you son of a bitch," I snarled, my voice cracking on the last word. "Don't leave me now. Not after everything."
The screen flickered again, and for a heartbeat, I thought—
"This is Lieutenant Kael Voss, and if you're hearing this, then we've already lost."
My breath caught. The voice was wrong. Too clean, too crisp, but it wasn't mine. It was hers. Her.
I staggered back as the comms crackled to life, the words pouring out in a flood that almost hurt. "They took the engines. I don't know how long before they find you, but you have to listen—don't trust the council. Don't trust anyone. Run. And if you survive…"
A chime interrupted her, sharp and shrill. I flinched, my hand flying to the sidearm at my hip. Not now. Not when the goddamn truth was finally slipping through my fingers.
The comms crackled again, and this time, it was my turn to beg. "C'mon, damn it! You can't just—you can't just leave me hanging like this!" My voice cracked, the words laced with something raw and broken. "I didn't come all this way to watch you die on me again, Lira!"
For a moment, there was only silence. Then her voice, softer now, almost tender. "You never could take a hint, could you?"
I closed my eyes against the burn behind my ribs. Not fair. Not fair. She'd always known how to twist the knife, hadn't she? Even here, even now.
"Kael…" Her voice dropped, and I could almost see her, standing in that damned lab, her auburn hair pulled back in a messy knot, eyes blazing with that stubborn fire of hers. "You have to get this transmission out. No matter what happens. Promised?"
I swallowed hard, the weight of the lie sitting heavy in my chest. "Yeah," I croaked. "Yeah, I promise."
The comms crackled again, and for a heartbeat, I thought—
Then it died.
Just like that. All the static, all the fury, all the damn hope she'd left behind, gone.
I slumped against the console, my breath coming in ragged gasps. "Gods," I whispered, the words barely audible over the hum of the failing systems. "You really did it, didn't you? You goddamn idiot."
But even as the words left my mouth, I knew it was true. She'd always been a handful. A pain in the ass. The kind of woman who made your heart turn to lead and your soul turn to ash with one sardonic smirk. And she'd done it.
She'd saved us all.
A shudder ran through me as the lights flickered again, casting wild shadows across the walls. I dragged a hand down my face, scrubbing at the exhaustion that felt like it was seeping into my very bones. Just a little longer. That's what she'd have said, wasn't it? Just a little longer, Kael.
A soft chime echoed through the cockpit, and for a heartbeat, I just stared at the console. My fingers itched to reach for it, to press that one last button and let the truth—her truth—ripple out into the void.
Because that's what we did. That's what we all did. We reached for each other, even when it hurt. Even when it was too late.
I exhaled sharply, the sound feeling like a prayer. "Alright," I murmured, my voice steady now. "Let's not make her a liar, huh?"
And with that, I pressed the final key.
|
mlx-community/gemma-3-1b-it-DQ | mlx-community | 2025-06-15T19:38:37Z | 131 | 0 | mlx | [
"mlx",
"safetensors",
"gemma3_text",
"text-generation",
"conversational",
"base_model:google/gemma-3-1b-it",
"base_model:quantized:google/gemma-3-1b-it",
"license:gemma",
"model-index",
"4-bit",
"region:us"
] | text-generation | 2025-06-12T03:30:12Z | ---
license: gemma
library_name: mlx
pipeline_tag: text-generation
extra_gated_heading: Access Gemma on Hugging Face
extra_gated_prompt: To access Gemma on Hugging Face, you’re required to review and
agree to Google’s usage license. To do this, please ensure you’re logged in to Hugging
Face and click below. Requests are processed immediately.
extra_gated_button_content: Acknowledge license
base_model: google/gemma-3-1b-it
tags:
- mlx
model-index:
- name: gemma-3-1b-it-DQ
results:
- task:
type: text-generation
dataset:
type: PIQA
name: PIQA
metrics:
- name: pass@1
type: pass@1
value: 0.75
verified: false
- task:
type: text-generation
dataset:
type: winogrande
name: winogrande
metrics:
- name: pass@1
type: pass@1
value: 0.60
verified: false
- task:
type: text-generation
dataset:
type: boolq
name: boolq
metrics:
- name: pass@1
type: pass@1
value: 0.73
verified: false
- task:
type: text-generation
dataset:
type: arc-c
name: arc-c
metrics:
- name: pass@1
type: pass@1
value: 0.35
verified: false
---
# mlx-community/gemma-3-1b-it-DQ
This model [mlx-community/gemma-3-1b-it-DQ](https://huggingface.co/mlx-community/gemma-3-1b-it-DQ) was
converted to MLX format from [google/gemma-3-1b-it](https://huggingface.co/google/gemma-3-1b-it)
using mlx-lm version **0.25.2**.
##### 2x faster and 2.4x less memory footprint than the dequantized model
## Use with mlx
```bash
pip install mlx-lm
```
```python
from mlx_lm import load, generate
model, tokenizer = load("mlx-community/gemma-3-1b-it-DQ")
prompt = "hello"
if tokenizer.chat_template is not None:
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
messages, add_generation_prompt=True
)
response = generate(model, tokenizer, prompt=prompt, verbose=True)
```
|
Mungert/Holo1-3B-GGUF | Mungert | 2025-06-15T19:38:35Z | 3,419 | 1 | transformers | [
"transformers",
"gguf",
"multimodal",
"action",
"agent",
"visual-document-retrieval",
"en",
"arxiv:2506.02865",
"arxiv:2401.13919",
"base_model:Qwen/Qwen2.5-VL-3B-Instruct",
"base_model:quantized:Qwen/Qwen2.5-VL-3B-Instruct",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | visual-document-retrieval | 2025-06-04T19:44:51Z | ---
base_model:
- Qwen/Qwen2.5-VL-3B-Instruct
language:
- en
library_name: transformers
license: other
license_name: other
pipeline_tag: visual-document-retrieval
tags:
- multimodal
- action
- agent
---
# <span style="color: #7FFF7F;">Holo1-3B GGUF Models</span>
This model is described in the paper [Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights](https://huggingface.co/papers/2506.02865). The project page can be found at [https://www.surferh.com](https://www.surferh.com).
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`71bdbdb5`](https://github.com/ggerganov/llama.cpp/commit/71bdbdb58757d508557e6d8b387f666cdfb25c5e).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Holo1-3B
## Model Description
Holo1 is an Action Vision-Language Model (VLM) developed by [HCompany](https://www.hcompany.ai/) for use in the Surfer-H web agent system. It is designed to interact with web interfaces like a human user.
As part of a broader agentic architecture, Holo1 acts as a policy, localizer, or validator, helping the agent understand and act in digital environments.
Trained on a mix of open-access, synthetic, and self-generated data, Holo1 enables state-of-the-art (SOTA) performance on the [WebVoyager](https://arxiv.org/pdf/2401.13919) benchmark, offering the best accuracy/cost tradeoff among current models.
It also excels in UI localization tasks such as [Screenspot](https://huggingface.co/datasets/rootsautomation/ScreenSpot), [Screenspot-V2](https://huggingface.co/datasets/HongxinLi/ScreenSpot_v2), [Screenspot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro), [GroundUI-Web](https://huggingface.co/datasets/agent-studio/GroundUI-1K), and our own newly introduced
benchmark [WebClick](https://huggingface.co/datasets/Hcompany/WebClick).
Holo1 is optimized for both accuracy and cost-efficiency, making it a strong open-source alternative to existing VLMs.
For more details, check our paper and our blog post.
- **Developed by:** [HCompany](https://www.hcompany.ai/)
- **Model type:** Action Vision-Language Model
- **Finetuned from model:** Qwen/Qwen2.5-VL-3B-Instruct
- **Paper:** https://arxiv.org/abs/2506.02865
- **Blog Post:** https://www.hcompany.ai/surfer-h
- **License:** https://huggingface.co/Hcompany/Holo1-3B/blob/main/LICENSE
## Results
### Surfer-H: Pareto-Optimal Performance on [WebVoyager](https://arxiv.org/pdf/2401.13919)
Surfer-H is designed to be flexible and modular. It is composed of three independent components:
- A Policy model that plans, decides, and drives the agent's behavior
- A Localizer model that sees and understands visual UIs to drive precise interactions
- A Validator model that checks whether the answer is valid
The agent thinks before acting, takes notes, and can retry if its answer is rejected. It can operate with different models for each module, allowing for tradeoffs between accuracy, speed, and cost.
We evaluated Surfer-H on the [WebVoyager](https://arxiv.org/pdf/2401.13919) benchmark: 643 real-world web tasks ranging from retrieving prices to finding news or scheduling events.
<div style="text-align: center;">
<img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/kO_4DlW_O45Wi7eK9-r8v.png" width="800"/>
</div>
We’ve tested multiple configurations, from GPT-4-powered agents to 100% open Holo1 setups. Among them, the fully Holo1-based agents offered the strongest tradeoff between accuracy and cost:
- Surfer-H + Holo1-7B: 92.2% accuracy at $0.13 per task
- Surfer-H + GPT-4.1: 92.0% at $0.54 per task
- Surfer-H + Holo1-3B: 89.7% at $0.11 per task
- Surfer-H + GPT-4.1-mini: 88.8% at $0.26 per task
This places Holo1-powered agents on the Pareto frontier, delivering the best accuracy per dollar.
Unlike other agents that rely on custom APIs or brittle wrappers, Surfer-H operates purely through the browser — just like a real user. Combined with Holo1, it becomes a powerful, general-purpose, cost-efficient web automation system.
### Holo1: State-of-the-Art UI Localization
A key skill for the real-world utility of our VLMs within agents is localization: the ability to identify precise
coordinates on a user interface (UI) to interact with to complete a task or follow an instruction. To assess
this capability, we evaluated our Holo1 models on several established localization benchmarks, including
[Screenspot](https://huggingface.co/datasets/rootsautomation/ScreenSpot), [Screenspot-V2](https://huggingface.co/datasets/HongxinLi/ScreenSpot_v2), [Screenspot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro), [GroundUI-Web](https://huggingface.co/datasets/agent-studio/GroundUI-1K), and our own newly introduced
benchmark [WebClick](https://huggingface.co/datasets/Hcompany/WebClick).
<div style="text-align: center;">
<img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/UutD2Meevd5Xw0_mhX2wK.png" width="600"/>
</div>
<div style="text-align: center;">
<img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/NhzkB8xnEQYMqiGxPnJSt.png" width="600"/>
</div>
## Get Started with the Model
We provide 2 spaces to experiment with Localization and Navigation:
- https://huggingface.co/spaces/Hcompany/Holo1-Navigation
- https://huggingface.co/spaces/Hcompany/Holo1-Localization
We provide starter code for the localization task: i.e. image + instruction -> click coordinates
We also provide code to reproduce screenspot evaluations: screenspot_eval.py
### Prepare model, processor
Holo1 models are based on Qwen2.5-VL architecture, which comes with transformers support. Here we provide a simple usage example.
You can load the model and the processor as follows:
```python
import json
import os
from typing import Any, Literal
from transformers import AutoModelForImageTextToText, AutoProcessor
# default: Load the model on the available device(s)
# We recommend enabling flash_attention_2 for better acceleration and memory saving.
model = AutoModelForImageTextToText.from_pretrained(
"Hcompany/Holo1-3B",
torch_dtype="auto",
# torch_dtype=torch.bfloat16,
# attn_implementation="flash_attention_2",
device_map="auto",
)
# default processor
processor = AutoProcessor.from_pretrained("Hcompany/Holo1-3B")
# The default range for the number of visual tokens per image in the model is 4-1280.
# You can set min_pixels and max_pixels according to your needs, such as a token range of 256-1280, to balance performance and cost.
# processor = AutoProcessor.from_pretrained(model_dir, min_pixels=min_pixels, max_pixels=max_pixels)
# Helper function to run inference
def run_inference(messages: list[dict[str, Any]]) -> str:
# Preparation for inference
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(
text=[text],
images=image,
padding=True,
return_tensors="pt",
)
inputs = inputs.to("cuda")
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
return processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
```
### Prepare image and instruction
WARNING: Holo1 is using absolute coordinates (number of pixels) and HuggingFace processor is doing image resize. To have matching coordinates, one needs to smart_resize the image.
```python
from PIL import Image
from transformers.models.qwen2_vl.image_processing_qwen2_vl import smart_resize
# Prepare image and instruction
image_url = "https://huggingface.co/Hcompany/Holo1-3B/resolve/main/calendar_example.jpg"
image = Image.open(requests.get(image_url, stream=True).raw)
# Resize the image so that predicted absolute coordinates match the size of the image.
image_processor = processor.image_processor
resized_height, resized_width = smart_resize(
image.height,
image.width,
factor=image_processor.patch_size * image_processor.merge_size,
min_pixels=image_processor.min_pixels,
max_pixels=image_processor.max_pixels,
)
image = image.resize(size=(resized_width, resized_height), resample=None) # type: ignore
```
### Navigation with Structured Output
```python
import json
from . import navigation
task = "Book a hotel in Paris on August 3rd for 3 nights"
prompt = navigation.get_navigation_prompt(task, image, step=1)
navigation_str = run_inference(prompt)[0]
navigation = navigation.NavigationStep(**json.loads(navigation_str))
print(navigation)
# Expected NavigationStep(note='', thought='I need to select the check-out date as August 3rd and then proceed to search for hotels.', action=ClickElementAction(action='click_element', element='August 3rd on the calendar', x=777, y=282))
```
### Localization with click(x, y)
```python
from . import localization
instruction = "Select July 14th as the check-out date"
prompt = localization.get_localization_prompt(image, instruction)
coordinates = run_inference(prompt)[0]
print(coordinates)
# Expected Click(352, 348)
```
### Localization with Structured Output
We trained Holo1 as an Action VLM with extensive use of json and tool calls. Therefore, it can be queried reliably with structured output:
```python
import json
from . import localization
instruction = "Select July 14th as the check-out date"
prompt = localization.get_localization_prompt_structured_output(image, instruction)
coordinates_structured_str = run_inference(prompt)[0]
coordinates_structured = localization.ClickAction(**json.loads(coordinates_structured_str))
print(coordinates_structured)
# Expected ClickAction(action='click', x=352, y=340)
```
## Citation
**BibTeX:**
```
@misc{andreux2025surferhmeetsholo1costefficient,
title={Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights},
author={Mathieu Andreux and Breno Baldas Skuk and Hamza Benchekroun and Emilien Biré and Antoine Bonnet and Riaz Bordie and Matthias Brunel and Pierre-Louis Cedoz and Antoine Chassang and Mickaël Chen and Alexandra D. Constantinou and Antoine d'Andigné and Hubert de La Jonquière and Aurélien Delfosse and Ludovic Denoyer and Alexis Deprez and Augustin Derupti and Michael Eickenberg and Mathïs Federico and Charles Kantor and Xavier Koegler and Yann Labbé and Matthew C. H. Lee and Erwan Le Jumeau de Kergaradec and Amir Mahla and Avshalom Manevich and Adrien Maret and Charles Masson and Rafaël Maurin and Arturo Mena and Philippe Modard and Axel Moyal and Axel Nguyen Kerbel and Julien Revelle and Mats L. Richter and María Santos and Laurent Sifre and Maxime Theillard and Marc Thibault and Louis Thiry and Léo Tronchon and Nicolas Usunier and Tony Wu},
year={2025},
eprint={2506.02865},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2506.02865},
}
``` |
Mungert/Nemotron-Research-Reasoning-Qwen-1.5B-GGUF | Mungert | 2025-06-15T19:38:27Z | 1,351 | 1 | null | [
"gguf",
"en",
"arxiv:2505.24864",
"base_model:deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
"base_model:quantized:deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-06-03T21:22:54Z | ---
license: cc-by-nc-4.0
language:
- en
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
---
# <span style="color: #7FFF7F;">Nemotron-Research-Reasoning-Qwen-1.5B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`ea1431b0`](https://github.com/ggerganov/llama.cpp/commit/ea1431b0fa3a8108aac1e0a94a13ccc4a749963e).
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Nemotron-Research-Reasoning-Qwen-1.5B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Nemotron-Research-Reasoning-Qwen-1.5B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Nemotron-Research-Reasoning-Qwen-1.5B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Nemotron-Research-Reasoning-Qwen-1.5B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Nemotron-Research-Reasoning-Qwen-1.5B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Nemotron-Research-Reasoning-Qwen-1.5B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Nemotron-Research-Reasoning-Qwen-1.5B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Nemotron-Research-Reasoning-Qwen-1.5B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Nemotron-Research-Reasoning-Qwen-1.5B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Nemotron-Research-Reasoning-Qwen-1.5B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Nemotron-Research-Reasoning-Qwen-1.5B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
<div align="center">
<span style="font-family: default; font-size: 1.5em;">Nemotron-Research-Reasoning-Qwen-1.5B</span>
<div>
🚀 The leading generalist reasoning model for cutting-edge research and development 🌟
</div>
</div>
## Introduction
Nemotron-Research-Reasoning-Qwen-1.5B is the world’s leading 1.5B open-weight model for complex reasoning tasks such as mathematical problems, coding challenges, scientific questions, and logic puzzles.
It is trained using the ProRL algorithm on a diverse and comprehensive set of datasets.
Our model has achieved impressive results, outperforming Deepseek’s 1.5B model by a large margin on a broad range of tasks, including math, coding, and GPQA.
This model is for research and development only.
## ProRL: Prolonged Reinforcement Learning
ProRL is designed to enable extended RL training periods that facilitate deeper exploration of reasoning strategies.
It enables more than 2k training steps and scale the training data across diverse tasks—from traditional math and code tasks to STEM problems, logical puzzles, and instruction following, which, we hypothesize, are crucial for generalization.
Based on Group Relative Policy Optimization (GRPO), ProRL introduces three key techniques:
1. Mitigating Entropy Collapse
2. Decoupled clip and dynamic sampling policy optimization (DAPO)
3. KL regularization and reference policy reset
Using ProRL, we developed the world's best 1.5B reasoning model that significantly outperforms its base model, DeepSeek-R1-1.5B, and matches or even surpasses the performance of DeepSeek-R1-7B across a diverse range of benchmarks.
Notably, compared to DeepSeek-R1-1.5B, we achieve average pass@1 improvements of 14.7\% on math benchmarks, 13.9\% on coding, 54.8\% on logic puzzles, 25.1\% on STEM reasoning, and 18.1\% on instruction-following tasks.
## Training Datasets
| Dataset | Link |
|---------------------------|-------------------------------------------------------------------------------------------|
| DeepScaleR-Preview-Dataset | [Link](https://huggingface.co/datasets/agentica-org/DeepScaleR-Preview-Dataset) |
| Eurus-2-RL-Data | [Link](https://huggingface.co/datasets/PRIME-RL/Eurus-2-RL-Data) |
| Reasoning-gym | [Link](https://github.com/open-thought/reasoning-gym) |
| IFEval | [Link](https://huggingface.co/datasets/nvidia/Llama-Nemotron-Post-Training-Dataset) |
| SCP-116K | [Link](https://huggingface.co/datasets/EricLu/SCP-116K) |
## Evaluation Results
Table 1: Performance (pass@1) comparison for benchmarks across Math domain.
| Model | AIME24 | AIME25 | AMC | Math | Minerva | Olympiad | Avg |
|-------------------------------|--------|--------|-------|-------|----------|----------|--------|
| DeepSeek-R1-Distill-Qwen-1.5B | 28.54 | 22.71 | 62.58 | 82.90 | 26.38 | 43.58 | 44.45 |
| DeepScaleR-1.5B | 40.21 | 31.46 | 73.04 | 89.36 | 41.57 | 51.63 | 54.54 |
| *DeepSeek-R1-Distill-Qwen-7B* | 53.54 | 40.83 | 82.83 | 93.68 | 50.60 | 57.66 | 63.19 |
| **Nemotron-Research-Reasoning-Qwen-1.5B** | **48.13** | **33.33** | **79.29** | **91.89** | **47.98** | **60.22** | **60.14** |
Table 2: Performance (pass@1) comparison across benchmarks for Code. We abbreviate benchmarks names for condecontests (cc), codeforces (cf), humanevalplus (human), and livecodebench (LCB).
| Model | apps | cc | cf | taco | human | LCB | Avg |
|-------------------------------|--------|--------|--------|--------|--------|--------|--------|
| DeepSeek-R1-Distill-Qwen-1.5B | 20.95 | 16.79 | 14.13 | 8.03 | 61.77 | 16.80 | 23.08 |
| DeepCoder-1.5B | 30.37 | 23.76 | 21.70 | 13.76 | 73.40 | 22.76 | 30.96 |
| *DeepSeek-R1-Distill-Qwen-7B* | 42.08 | 32.76 | 33.08 | 19.08 | 83.32 | 38.04 | 41.39 |
| **Nemotron-Research-Reasoning-Qwen-1.5B** | **41.99** | **31.80** | **34.50** | **20.81** | 72.05 | **23.81** | **37.49** |
Table 3: Performance comparison on STEM reasoning (GPQA Diamond), instruction following (IFEval), and logic puzzles (Reasoning Gym) tasks. We also present results on OOD tasks: acre, boxnet, and game_of_life_halting (game).
| Model | GPQA | IFEval | Reasoning | acre | boxnet | game |
|-------------------------------|--------|--------|-----------|--------|--------|--------|
| DeepSeek-R1-Distill-Qwen-1.5B | 15.86 | 44.05 | 4.24 | 5.99 | 0.00 | 3.49 |
| *DeepSeek-R1-Distill-Qwen-7B* | 35.44 | 58.01 | 28.55 | 20.21 | 1.71 | 12.94 |
| **Nemotron-Research-Reasoning-Qwen-1.5B** | **41.78** | **66.02** | **59.06** | **58.57** | **7.91** | **52.29** |
## License/Terms of Use
cc-by-nc-4.0
## Ethical Considerations
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
## Citation
If you find our dataset helpful, please cite the following [paper](https://arxiv.org/abs/2505.24864):
```
@article{liu2025prorl,
author = {Mingjie Liu, Shizhe Diao, Ximing Lu, Jian Hu, Xin Dong, Yejin Choi, Jan Kautz, Yi Dong},
title={ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models},
journal = {arXiv preprint},
year = {2025},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url={https://arxiv.org/abs/2505.24864},
}
``` |
Mungert/Qwen3-16B-A3B-GGUF | Mungert | 2025-06-15T19:38:08Z | 3,213 | 4 | null | [
"gguf",
"unsloth",
"qwen3",
"qwen",
"base_model:kalomaze/Qwen3-16B-A3B",
"base_model:quantized:kalomaze/Qwen3-16B-A3B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-05-22T11:04:29Z | ---
tags:
- unsloth
- qwen3
- qwen
license: apache-2.0
base_model:
- kalomaze/Qwen3-16B-A3B
---
# <span style="color: #7FFF7F;">Qwen3-16B-A3B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`f5cd27b7`](https://github.com/ggerganov/llama.cpp/commit/f5cd27b71da3ac375a04a41643d14fc779a8057b).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Qwen3-16B-A3B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Qwen3-16B-A3B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Qwen3-16B-A3B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Qwen3-16B-A3B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Qwen3-16B-A3B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Qwen3-16B-A3B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Qwen3-16B-A3B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Qwen3-16B-A3B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Qwen3-16B-A3B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Qwen3-16B-A3B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Qwen3-16B-A3B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Qwen3-16B-A3B
Qwen3-16B-A3B is a rendition of Qwen3-30B-A3B by [kalomaze](https://huggingface.co/kalomaze/Qwen3-16B-A3B).
A man-made horror beyond your comprehension.
But no, seriously, this is my experiment to:
- measure the probability that any given expert will activate (over my personal set of fairly diverse calibration data), per layer
- prune 64/128 of the least used experts per layer (with reordered router and indexing per layer)
It can still write semi-coherently without any additional training or distillation done on top of it from the original 30b MoE.
The .txt files with the original measurements are provided in the repo along with the exported weights.
Custom testing to measure the experts was done on a hacked version of vllm, and then I made a bespoke script to selectively export the weights according to the measurements. |
Mungert/LiveCC-7B-Instruct-GGUF | Mungert | 2025-06-15T19:37:37Z | 1,832 | 1 | null | [
"gguf",
"qwen_vl",
"video",
"real-time",
"multimodal",
"LLM",
"en",
"dataset:chenjoya/Live-CC-5M",
"dataset:chenjoya/Live-WhisperX-526K",
"dataset:lmms-lab/LLaVA-Video-178K",
"arxiv:2504.16030",
"base_model:Qwen/Qwen2-VL-7B",
"base_model:quantized:Qwen/Qwen2-VL-7B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-04-27T00:50:10Z | ---
license: apache-2.0
datasets:
- chenjoya/Live-CC-5M
- chenjoya/Live-WhisperX-526K
- lmms-lab/LLaVA-Video-178K
language:
- en
base_model:
- Qwen/Qwen2-VL-7B
tags:
- qwen_vl
- video
- real-time
- multimodal
- LLM
---
# <span style="color: #7FFF7F;">LiveCC-7B-Instruct GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`e291450`](https://github.com/ggerganov/llama.cpp/commit/e291450b7602d7a36239e4ceeece37625f838373).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `LiveCC-7B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `LiveCC-7B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `LiveCC-7B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `LiveCC-7B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `LiveCC-7B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `LiveCC-7B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `LiveCC-7B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `LiveCC-7B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `LiveCC-7B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `LiveCC-7B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `LiveCC-7B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# LiveCC-7B-Instruct
## Introduction
We introduce LiveCC, the first video LLM capable of real-time commentary, trained with a novel video-ASR streaming method, SOTA on both streaming and offline benchmarks.
- Project Page: https://showlab.github.io/livecc
> [!Important]
> This is the SFT model. The base model is at [LiveCC-7B-Base](https://huggingface.co/chenjoya/LiveCC-7B-Base).
## Training with Streaming Frame-Words Paradigm
## Quickstart
### Gradio Demo
Please refer to https://github.com/showlab/livecc:
### Hands-on
Like qwen-vl-utils, we offer a toolkit to help you handle various types of visual input more conveniently, **especially on video streaming inputs**. You can install it using the following command:
```bash
pip install qwen-vl-utils livecc-utils liger_kernel
```
Here we show a code snippet to show you how to do **real-time video commentary** with `transformers` and the above utils:
```python
import functools, torch, os, tqdm
from liger_kernel.transformers import apply_liger_kernel_to_qwen2_vl
apply_liger_kernel_to_qwen2_vl() # important. our model is trained with this. keep consistency
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, LogitsProcessor, logging
from livecc_utils import prepare_multiturn_multimodal_inputs_for_generation, get_smart_resized_clip, get_smart_resized_video_reader
from qwen_vl_utils import process_vision_info
class LiveCCDemoInfer:
fps = 2
initial_fps_frames = 6
streaming_fps_frames = 2
initial_time_interval = initial_fps_frames / fps
streaming_time_interval = streaming_fps_frames / fps
frame_time_interval = 1 / fps
def __init__(self, model_path: str = None, device_id: int = 0):
self.model = Qwen2VLForConditionalGeneration.from_pretrained(
model_path, torch_dtype="auto",
device_map=f'cuda:{device_id}',
attn_implementation='flash_attention_2'
)
self.processor = AutoProcessor.from_pretrained(model_path, use_fast=False)
self.model.prepare_inputs_for_generation = functools.partial(prepare_multiturn_multimodal_inputs_for_generation, self.model)
message = {
"role": "user",
"content": [
{"type": "text", "text": 'livecc'},
]
}
texts = self.processor.apply_chat_template([message], tokenize=False)
self.system_prompt_offset = texts.index('<|im_start|>user')
self._cached_video_readers_with_hw = {}
def live_cc(
self,
query: str,
state: dict,
max_pixels: int = 384 * 28 * 28,
default_query: str = 'Please describe the video.',
do_sample: bool = True,
repetition_penalty: float = 1.05,
**kwargs,
):
"""
state: dict, (maybe) with keys:
video_path: str, video path
video_timestamp: float, current video timestamp
last_timestamp: float, last processed video timestamp
last_video_pts_index: int, last processed video frame index
video_pts: np.ndarray, video pts
last_history: list, last processed history
past_key_values: llm past_key_values
past_ids: past generated ids
"""
# 1. preparation: video_reader, and last processing info
video_timestamp, last_timestamp = state.get('video_timestamp', 0), state.get('last_timestamp', -1 / self.fps)
video_path = state['video_path']
if video_path not in self._cached_video_readers_with_hw:
self._cached_video_readers_with_hw[video_path] = get_smart_resized_video_reader(video_path, max_pixels)
video_reader = self._cached_video_readers_with_hw[video_path][0]
video_reader.get_frame_timestamp(0)
state['video_pts'] = torch.from_numpy(video_reader._frame_pts[:, 1])
state['last_video_pts_index'] = -1
video_pts = state['video_pts']
if last_timestamp + self.frame_time_interval > video_pts[-1]:
state['video_end'] = True
return
video_reader, resized_height, resized_width = self._cached_video_readers_with_hw[video_path]
last_video_pts_index = state['last_video_pts_index']
# 2. which frames will be processed
initialized = last_timestamp >= 0
if not initialized:
video_timestamp = max(video_timestamp, self.initial_time_interval)
if video_timestamp <= last_timestamp + self.frame_time_interval:
return
timestamps = torch.arange(last_timestamp + self.frame_time_interval, video_timestamp, self.frame_time_interval) # add compensation
# 3. fetch frames in required timestamps
clip, clip_timestamps, clip_idxs = get_smart_resized_clip(video_reader, resized_height, resized_width, timestamps, video_pts, video_pts_index_from=last_video_pts_index+1)
state['last_video_pts_index'] = clip_idxs[-1]
state['last_timestamp'] = clip_timestamps[-1]
# 4. organize to interleave frames
interleave_clips, interleave_timestamps = [], []
if not initialized:
interleave_clips.append(clip[:self.initial_fps_frames])
interleave_timestamps.append(clip_timestamps[:self.initial_fps_frames])
clip = clip[self.initial_fps_frames:]
clip_timestamps = clip_timestamps[self.initial_fps_frames:]
if len(clip) > 0:
interleave_clips.extend(list(clip.split(self.streaming_fps_frames)))
interleave_timestamps.extend(list(clip_timestamps.split(self.streaming_fps_frames)))
# 5. make conversation and send to model
for clip, timestamps in zip(interleave_clips, interleave_timestamps):
start_timestamp, stop_timestamp = timestamps[0].item(), timestamps[-1].item() + self.frame_time_interval
message = {
"role": "user",
"content": [
{"type": "text", "text": f'Time={start_timestamp:.1f}-{stop_timestamp:.1f}s'},
{"type": "video", "video": clip}
]
}
if not query and not state.get('query', None):
query = default_query
print(f'No query provided, use default_query={default_query}')
if query and state.get('query', None) != query:
message['content'].append({"type": "text", "text": query})
state['query'] = query
texts = self.processor.apply_chat_template([message], tokenize=False, add_generation_prompt=True, return_tensors='pt')
past_ids = state.get('past_ids', None)
if past_ids is not None:
texts = '<|im_end|>\n' + texts[self.system_prompt_offset:]
inputs = self.processor(
text=texts,
images=None,
videos=[clip],
return_tensors="pt",
return_attention_mask=False
)
inputs.to('cuda')
if past_ids is not None:
inputs['input_ids'] = torch.cat([past_ids, inputs.input_ids], dim=1)
outputs = self.model.generate(
**inputs, past_key_values=state.get('past_key_values', None),
return_dict_in_generate=True, do_sample=do_sample,
repetition_penalty=repetition_penalty,
)
state['past_key_values'] = outputs.past_key_values
state['past_ids'] = outputs.sequences[:, :-1]
yield (start_timestamp, stop_timestamp), self.processor.decode(outputs.sequences[0, inputs.input_ids.size(1):], skip_special_tokens=True), state
model_path = 'chenjoya/LiveCC-7B-Instruct'
# download a test video at: https://github.com/showlab/livecc/blob/main/demo/sources/howto_fix_laptop_mute_1080p.mp4
video_path = "demo/sources/howto_fix_laptop_mute_1080p.mp4"
query = "Please describe the video."
infer = LiveCCDemoInfer(model_path=model_path)
state = {'video_path': video_path}
commentaries = []
t = 0
for t in range(31):
state['video_timestamp'] = t
for (start_t, stop_t), response, state in infer.live_cc(
query=query, state=state,
max_pixels = 384 * 28 * 28, repetition_penalty=1.05,
streaming_eos_base_threshold=0.0, streaming_eos_threshold_step=0
):
print(f'{start_t}s-{stop_t}s: {response}')
commentaries.append([start_t, stop_t, response])
if state.get('video_end', False):
break
t += 1
```
Here we show a code snippet to show you how to do **common video (multi-turn) qa** with `transformers` and the above utils:
```python
import functools, torch
from liger_kernel.transformers import apply_liger_kernel_to_qwen2_vl
apply_liger_kernel_to_qwen2_vl() # important. our model is trained with this. keep consistency
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, LogitsProcessor, logging
from livecc_utils import prepare_multiturn_multimodal_inputs_for_generation, get_smart_resized_clip, get_smart_resized_video_reader
from qwen_vl_utils import process_vision_info
class LiveCCDemoInfer:
fps = 2
initial_fps_frames = 6
streaming_fps_frames = 2
initial_time_interval = initial_fps_frames / fps
streaming_time_interval = streaming_fps_frames / fps
frame_time_interval = 1 / fps
def __init__(self, model_path: str = None, device: str = 'cuda'):
self.model = Qwen2VLForConditionalGeneration.from_pretrained(
model_path, torch_dtype="auto",
device_map=device,
attn_implementation='flash_attention_2'
)
self.processor = AutoProcessor.from_pretrained(model_path, use_fast=False)
self.streaming_eos_token_id = self.processor.tokenizer(' ...').input_ids[-1]
self.model.prepare_inputs_for_generation = functools.partial(prepare_multiturn_multimodal_inputs_for_generation, self.model)
message = {
"role": "user",
"content": [
{"type": "text", "text": 'livecc'},
]
}
texts = self.processor.apply_chat_template([message], tokenize=False)
self.system_prompt_offset = texts.index('<|im_start|>user')
def video_qa(
self,
message: str,
state: dict,
do_sample: bool = True,
repetition_penalty: float = 1.05,
**kwargs,
):
"""
state: dict, (maybe) with keys:
video_path: str, video path
video_timestamp: float, current video timestamp
last_timestamp: float, last processed video timestamp
last_video_pts_index: int, last processed video frame index
video_pts: np.ndarray, video pts
last_history: list, last processed history
past_key_values: llm past_key_values
past_ids: past generated ids
"""
video_path = state.get('video_path', None)
conversation = []
past_ids = state.get('past_ids', None)
content = [{"type": "text", "text": message}]
if past_ids is None and video_path: # only use once
content.insert(0, {"type": "video", "video": video_path})
conversation.append({"role": "user", "content": content})
image_inputs, video_inputs = process_vision_info(conversation)
texts = self.processor.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True, return_tensors='pt')
if past_ids is not None:
texts = '<|im_end|>\n' + texts[self.system_prompt_offset:]
inputs = self.processor(
text=texts,
images=image_inputs,
videos=video_inputs,
return_tensors="pt",
return_attention_mask=False
)
inputs.to(self.model.device)
if past_ids is not None:
inputs['input_ids'] = torch.cat([past_ids, inputs.input_ids], dim=1)
outputs = self.model.generate(
**inputs, past_key_values=state.get('past_key_values', None),
return_dict_in_generate=True, do_sample=do_sample,
repetition_penalty=repetition_penalty,
max_new_tokens=512,
)
state['past_key_values'] = outputs.past_key_values
state['past_ids'] = outputs.sequences[:, :-1]
response = self.processor.decode(outputs.sequences[0, inputs.input_ids.size(1):], skip_special_tokens=True)
return response, state
model_path = 'chenjoya/LiveCC-7B-Instruct'
# download a test video at: https://github.com/showlab/livecc/blob/main/demo/sources/howto_fix_laptop_mute_1080p.mp4
video_path = "demo/sources/howto_fix_laptop_mute_1080p.mp4"
infer = LiveCCDemoInfer(model_path=model_path)
state = {'video_path': video_path}
# first round
query1 = 'What is the video?'
response1, state = infer.video_qa(message=query1, state=state)
print(f'Q1: {query1}\nA1: {response1}')
# second round
query2 = 'How do you know that?'
response2, state = infer.video_qa(message=query2, state=state)
print(f'Q2: {query2}\nA2: {response2}')
```
## Performance
## Limitations
- This model is finetuned on LiveCC-7B-Base, which is starting from Qwen2-VL-7B-Base, so it may have limitations mentioned in https://huggingface.co/Qwen/Qwen2-VL-7B.
- When performing real-time video commentary, it may appear collapse --- e.g., repeat pattern. If you encounter this situation, try to adjust repetition_penalty, streaming_eos_base_threshold, and streaming_eos_threshold_step.
- This model only has a context window of 32768. Using more visual tokens per frame (e.g. 768 * 28 * 28) will have better performance, but will shorten the working duration.
These limitations serve as ongoing directions for model optimization and improvement, and we are committed to continually enhancing the model's performance and scope of application.
## Citation
If you find our work helpful, feel free to give us a cite.
```
@article{livecc,
author = {Joya Chen and Ziyun Zeng and Yiqi Lin and Wei Li and Zejun Ma and Mike Zheng Shou},
title = {LiveCC: Learning Video LLM with Streaming Speech Transcription at Scale},
journal = {arXiv preprint arXiv:2504.16030}
year = {2025},
}
``` |
Mungert/GLM-Z1-9B-0414-GGUF | Mungert | 2025-06-15T19:37:27Z | 1,000 | 5 | transformers | [
"transformers",
"gguf",
"text-generation",
"zh",
"en",
"arxiv:2406.12793",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-04-26T06:57:42Z | ---
license: mit
language:
- zh
- en
pipeline_tag: text-generation
library_name: transformers
---
# <span style="color: #7FFF7F;">GLM-Z1-9B-0414 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`e291450`](https://github.com/ggerganov/llama.cpp/commit/e291450b7602d7a36239e4ceeece37625f838373).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `GLM-Z1-9B-0414-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `GLM-Z1-9B-0414-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `GLM-Z1-9B-0414-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `GLM-Z1-9B-0414-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `GLM-Z1-9B-0414-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `GLM-Z1-9B-0414-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `GLM-Z1-9B-0414-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `GLM-Z1-9B-0414-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `GLM-Z1-9B-0414-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `GLM-Z1-9B-0414-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `GLM-Z1-9B-0414-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# GLM-4-Z1-9B-0414
## Introduction
The GLM family welcomes a new generation of open-source models, the **GLM-4-32B-0414** series, featuring 32 billion parameters. Its performance is comparable to OpenAI's GPT series and DeepSeek's V3/R1 series, and it supports very user-friendly local deployment features. GLM-4-32B-Base-0414 was pre-trained on 15T of high-quality data, including a large amount of reasoning-type synthetic data, laying the foundation for subsequent reinforcement learning extensions. In the post-training stage, in addition to human preference alignment for dialogue scenarios, we also enhanced the model's performance in instruction following, engineering code, and function calling using techniques such as rejection sampling and reinforcement learning, strengthening the atomic capabilities required for agent tasks. GLM-4-32B-0414 achieves good results in areas such as engineering code, Artifact generation, function calling, search-based Q&A, and report generation. Some benchmarks even rival larger models like GPT-4o and DeepSeek-V3-0324 (671B).
**GLM-Z1-32B-0414** is a reasoning model with **deep thinking capabilities**. This was developed based on GLM-4-32B-0414 through cold start and extended reinforcement learning, as well as further training of the model on tasks involving mathematics, code, and logic. Compared to the base model, GLM-Z1-32B-0414 significantly improves mathematical abilities and the capability to solve complex tasks. During the training process, we also introduced general reinforcement learning based on pairwise ranking feedback, further enhancing the model's general capabilities.
**GLM-Z1-Rumination-32B-0414** is a deep reasoning model with **rumination capabilities** (benchmarked against OpenAI's Deep Research). Unlike typical deep thinking models, the rumination model employs longer periods of deep thought to solve more open-ended and complex problems (e.g., writing a comparative analysis of AI development in two cities and their future development plans). The rumination model integrates search tools during its deep thinking process to handle complex tasks and is trained by utilizing multiple rule-based rewards to guide and extend end-to-end reinforcement learning. Z1-Rumination shows significant improvements in research-style writing and complex retrieval tasks.
Finally, **GLM-Z1-9B-0414** is a surprise. We employed the aforementioned series of techniques to train a 9B small-sized model that maintains the open-source tradition. Despite its smaller scale, GLM-Z1-9B-0414 still exhibits excellent capabilities in mathematical reasoning and general tasks. Its overall performance is already at a leading level among open-source models of the same size. Especially in resource-constrained scenarios, this model achieves an excellent balance between efficiency and effectiveness, providing a powerful option for users seeking lightweight deployment.
## Performance
<p align="center">
<img width="100%" src="https://raw.githubusercontent.com/THUDM/GLM-4/refs/heads/main/resources/Bench-Z1-32B.png">
</p>
<p align="center">
<img width="100%" src="https://raw.githubusercontent.com/THUDM/GLM-4/refs/heads/main/resources/Bench-Z1-9B.png">
</p>
## Model Usage Guidelines
### I. Sampling Parameters
| Parameter | Recommended Value | Description |
| ------------ | ----------------- | -------------------------------------------- |
| temperature | **0.6** | Balances creativity and stability |
| top_p | **0.95** | Cumulative probability threshold for sampling|
| top_k | **40** | Filters out rare tokens while maintaining diversity |
| max_new_tokens | **30000** | Leaves enough tokens for thinking |
### II. Enforced Thinking
- Add \<think\>\n to the **first line**: Ensures the model thinks before responding
- When using `chat_template.jinja`, the prompt is automatically injected to enforce this behavior
### III. Dialogue History Trimming
- Retain only the **final user-visible reply**.
Hidden thinking content should **not** be saved to history to reduce interference—this is already implemented in `chat_template.jinja`
### IV. Handling Long Contexts (YaRN)
- When input length exceeds **8,192 tokens**, consider enabling YaRN (Rope Scaling)
- In supported frameworks, add the following snippet to `config.json`:
```json
"rope_scaling": {
"type": "yarn",
"factor": 4.0,
"original_max_position_embeddings": 32768
}
```
- **Static YaRN** applies uniformly to all text. It may slightly degrade performance on short texts, so enable as needed.
## Inference Code
Make Sure Using `transforemrs>=4.51.3`.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL_PATH = "THUDM/GLM-4-Z1-9B-0414"
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, device_map="auto")
message = [{"role": "user", "content": "Let a, b be positive real numbers such that ab = a + b + 3. Determine the range of possible values for a + b."}]
inputs = tokenizer.apply_chat_template(
message,
return_tensors="pt",
add_generation_prompt=True,
return_dict=True,
).to(model.device)
generate_kwargs = {
"input_ids": inputs["input_ids"],
"attention_mask": inputs["attention_mask"],
"max_new_tokens": 4096,
"do_sample": False,
}
out = model.generate(**generate_kwargs)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```
## Citations
If you find our work useful, please consider citing the following paper.
```
@misc{glm2024chatglm,
title={ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools},
author={Team GLM and Aohan Zeng and Bin Xu and Bowen Wang and Chenhui Zhang and Da Yin and Diego Rojas and Guanyu Feng and Hanlin Zhao and Hanyu Lai and Hao Yu and Hongning Wang and Jiadai Sun and Jiajie Zhang and Jiale Cheng and Jiayi Gui and Jie Tang and Jing Zhang and Juanzi Li and Lei Zhao and Lindong Wu and Lucen Zhong and Mingdao Liu and Minlie Huang and Peng Zhang and Qinkai Zheng and Rui Lu and Shuaiqi Duan and Shudan Zhang and Shulin Cao and Shuxun Yang and Weng Lam Tam and Wenyi Zhao and Xiao Liu and Xiao Xia and Xiaohan Zhang and Xiaotao Gu and Xin Lv and Xinghan Liu and Xinyi Liu and Xinyue Yang and Xixuan Song and Xunkai Zhang and Yifan An and Yifan Xu and Yilin Niu and Yuantao Yang and Yueyan Li and Yushi Bai and Yuxiao Dong and Zehan Qi and Zhaoyu Wang and Zhen Yang and Zhengxiao Du and Zhenyu Hou and Zihan Wang},
year={2024},
eprint={2406.12793},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
``` |
ALYTV/Qwen2.5-Coder-7B-mlx-6Bit | ALYTV | 2025-06-15T19:37:22Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"code",
"qwen",
"qwen-coder",
"codeqwen",
"mlx",
"mlx-my-repo",
"conversational",
"en",
"base_model:Qwen/Qwen2.5-Coder-7B",
"base_model:quantized:Qwen/Qwen2.5-Coder-7B",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"6-bit",
"region:us"
] | text-generation | 2025-06-15T19:36:56Z | ---
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen2.5-Coder-7B/blob/main/LICENSE
language:
- en
base_model: Qwen/Qwen2.5-Coder-7B
pipeline_tag: text-generation
library_name: transformers
tags:
- code
- qwen
- qwen-coder
- codeqwen
- mlx
- mlx-my-repo
---
# ALYTV/Qwen2.5-Coder-7B-mlx-6Bit
The Model [ALYTV/Qwen2.5-Coder-7B-mlx-6Bit](https://huggingface.co/ALYTV/Qwen2.5-Coder-7B-mlx-6Bit) was converted to MLX format from [Qwen/Qwen2.5-Coder-7B](https://huggingface.co/Qwen/Qwen2.5-Coder-7B) using mlx-lm version **0.22.3**.
## Use with mlx
```bash
pip install mlx-lm
```
```python
from mlx_lm import load, generate
model, tokenizer = load("ALYTV/Qwen2.5-Coder-7B-mlx-6Bit")
prompt="hello"
if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
response = generate(model, tokenizer, prompt=prompt, verbose=True)
```
|
Mungert/watt-tool-8B-GGUF | Mungert | 2025-06-15T19:37:09Z | 1,358 | 6 | null | [
"gguf",
"function-calling",
"tool-use",
"llama",
"bfcl",
"en",
"arxiv:2406.14868",
"base_model:meta-llama/Llama-3.1-8B-Instruct",
"base_model:quantized:meta-llama/Llama-3.1-8B-Instruct",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-04-15T01:43:23Z | ---
license: apache-2.0
language:
- en
base_model:
- meta-llama/Llama-3.1-8B-Instruct
tags:
- function-calling
- tool-use
- llama
- bfcl
---
# <span style="color: #7FFF7F;">watt-tool-8B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `watt-tool-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `watt-tool-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `watt-tool-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `watt-tool-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `watt-tool-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `watt-tool-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `watt-tool-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `watt-tool-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `watt-tool-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `watt-tool-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `watt-tool-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# watt-tool-8B
watt-tool-8B is a fine-tuned language model based on LLaMa-3.1-8B-Instruct, optimized for tool usage and multi-turn dialogue. It achieves state-of-the-art performance on the Berkeley Function-Calling Leaderboard (BFCL).
## Model Description
This model is specifically designed to excel at complex tool usage scenarios that require multi-turn interactions, making it ideal for empowering platforms like [Lupan](https://lupan.watt.chat), an AI-powered workflow building tool. By leveraging a carefully curated and optimized dataset, watt-tool-8B demonstrates superior capabilities in understanding user requests, selecting appropriate tools, and effectively utilizing them across multiple turns of conversation.
Target Application: AI Workflow Building as in [https://lupan.watt.chat/](https://lupan.watt.chat/) and [Coze](https://www.coze.com/).
## Key Features
* **Enhanced Tool Usage:** Fine-tuned for precise and efficient tool selection and execution.
* **Multi-Turn Dialogue:** Optimized for maintaining context and effectively utilizing tools across multiple turns of conversation, enabling more complex task completion.
* **State-of-the-Art Performance:** Achieves top performance on the BFCL, demonstrating its capabilities in function calling and tool usage.
## Training Methodology
watt-tool-8B is trained using supervised fine-tuning on a specialized dataset designed for tool usage and multi-turn dialogue. We use CoT techniques to synthesize high-quality multi-turn dialogue data.
The training process is inspired by the principles outlined in the paper: ["Direct Multi-Turn Preference Optimization for Language Agents"](https://arxiv.org/abs/2406.14868).
We use SFT and DMPO to further enhance the model's performance in multi-turn agent tasks.
## How to Use
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "watt-ai/watt-tool-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype='auto', device_map="auto")
# Example usage (adapt as needed for your specific tool usage scenario)
"""You are an expert in composing functions. You are given a question and a set of possible functions. Based on the question, you will need to make one or more function/tool calls to achieve the purpose.
If none of the function can be used, point it out. If the given question lacks the parameters required by the function, also point it out.
You should only return the function call in tools call sections.
If you decide to invoke any of the function(s), you MUST put it in the format of [func_name1(params_name1=params_value1, params_name2=params_value2...), func_name2(params)]
You SHOULD NOT include any other text in the response.
Here is a list of functions in JSON format that you can invoke.\n{functions}\n
"""
# User query
query = "Find me the sales growth rate for company XYZ for the last 3 years and also the interest coverage ratio for the same duration."
tools = [
{
"name": "financial_ratios.interest_coverage", "description": "Calculate a company's interest coverage ratio given the company name and duration",
"arguments": {
"type": "dict",
"properties": {
"company_name": {
"type": "string",
"description": "The name of the company."
},
"years": {
"type": "integer",
"description": "Number of past years to calculate the ratio."
}
},
"required": ["company_name", "years"]
}
},
{
"name": "sales_growth.calculate",
"description": "Calculate a company's sales growth rate given the company name and duration",
"arguments": {
"type": "dict",
"properties": {
"company": {
"type": "string",
"description": "The company that you want to get the sales growth rate for."
},
"years": {
"type": "integer",
"description": "Number of past years for which to calculate the sales growth rate."
}
},
"required": ["company", "years"]
}
},
{
"name": "weather_forecast",
"description": "Retrieve a weather forecast for a specific location and time frame.",
"arguments": {
"type": "dict",
"properties": {
"location": {
"type": "string",
"description": "The city that you want to get the weather for."
},
"days": {
"type": "integer",
"description": "Number of days for the forecast."
}
},
"required": ["location", "days"]
}
}
]
messages = [
{'role': 'system', 'content': system_prompt.format(functions=tools)},
{'role': 'user', 'content': query}
]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)) |
Mungert/DeepCoder-14B-Preview-GGUF | Mungert | 2025-06-15T19:37:06Z | 1,424 | 9 | transformers | [
"transformers",
"gguf",
"text-generation",
"en",
"dataset:PrimeIntellect/verifiable-coding-problems",
"dataset:likaixin/TACO-verified",
"dataset:livecodebench/code_generation_lite",
"base_model:deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
"base_model:quantized:deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-04-11T03:58:56Z | ---
license: mit
library_name: transformers
datasets:
- PrimeIntellect/verifiable-coding-problems
- likaixin/TACO-verified
- livecodebench/code_generation_lite
language:
- en
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">DeepCoder-14B-Preview GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `DeepCoder-14B-Preview-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `DeepCoder-14B-Preview-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `DeepCoder-14B-Preview-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `DeepCoder-14B-Preview-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `DeepCoder-14B-Preview-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `DeepCoder-14B-Preview-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `DeepCoder-14B-Preview-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `DeepCoder-14B-Preview-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `DeepCoder-14B-Preview-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `DeepCoder-14B-Preview-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `DeepCoder-14B-Preview-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Quantum Network Monitor Agent](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
<div align="center">
<span style="font-family: default; font-size: 1.5em;">DeepCoder-14B-Preview</span>
<div>
🚀 Democratizing Reinforcement Learning for LLMs (RLLM) 🌟
</div>
</div>
<br>
<div align="center" style="line-height: 1;">
<a href="https://github.com/agentica-project/rllm" style="margin: 2px;">
<img alt="Code" src="https://img.shields.io/badge/RLLM-000000?style=for-the-badge&logo=github&logoColor=000&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://pretty-radio-b75.notion.site/DeepCoder-A-Fully-Open-Source-14B-Coder-at-O3-mini-Level-1cf81902c14680b3bee5eb349a512a51" target="_blank" style="margin: 2px;">
<img alt="Blog" src="https://img.shields.io/badge/Notion-%23000000.svg?style=for-the-badge&logo=notion&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://x.com/Agentica_" style="margin: 2px;">
<img alt="X.ai" src="https://img.shields.io/badge/Agentica-white?style=for-the-badge&logo=X&logoColor=000&color=000&labelColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://huggingface.co/agentica-org" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/Agentica-fcd022?style=for-the-badge&logo=huggingface&logoColor=000&labelColor" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://www.together.ai" style="margin: 2px;">
<img alt="Together AI" src="https://img.shields.io/badge/-Together_AI%20-white?style=for-the-badge&logo=data%3Aimage%2Fpng%3Bbase64%2CiVBORw0KGgoAAAANSUhEUgAAAUAAAAFACAMAAAD6TlWYAAAC7lBMVEUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAPb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8AAAAPb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8Pb%2F8AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADIBDt6AAAA%2BnRSTlMAAiQEKgcdKQwiHBMUzrtSUEmjhmZGH96yv8n1ey7nL3y1U%2FZfCaIo1WFg1NrcsHYrA2%2Fv80J%2BMeilnpefqKw%2B64%2BQlSbYZGVnBGkCV%2BxW8XJube6WJ9kZF9bSzBALRynPQfLhIjvwyBEAXOTLp3o%2FJA9Y9%2F7%2F9FEKDhIVFo4GHkVzjGz8icrHzY39iHR1i0M8Jj14LLZUvb7DxMXGoQEFeQcgSBOHaPvm4uOdRLMMqcDTLbcII0sNuVn4TKaRd6RKIeDd37Svra6xuLpaW17lXUAlHh8WGxUPIS4JGQoFECMsBg4gFwsRJRIrCC0oAycaFC8NMDIzMRgBsVt9rwAAD25JREFUeNrs3QVzG0kWB%2FA3ikHhZeYwk3LMbF7GcBasOGw9hb3MzLyKw8zMzMx2rsokhySNY2mmR1N4xXV3a7sHuzWu%2BX2Ef3XPG%2Br3wOVyuVwul8vlcrlcLpfL5XK5dOlXOHTIvLnb27Xd%2FasBvrt9A%2B7r1bbdTTffcmuXwhzgTYwk6q%2BHr2RWlcclRYqXV2VeCV%2Bvr4mIkCJKZ83uc9NLC0fMD%2BD%2FCswfMfLtzh%2FeelsJcKJW19SG66KSTP6fLEXrwrU11Srw5Z8zbuzePcUBbFyg%2BPY7Pv%2Bs0A%2Bsid7ayiqFNEWp8iS9Ir%2F0Cl957bkRAaQLFLz15sBBfpbpJc7FJKKFFGuV4JJh6N573g6idr7vP%2F8iC9iI1NZJRDupLnlRBbaW3XjTfQHUJ3D8d68MBtsJiTNRold5uEYAdibkHgqiESMefGi9zfFVeCRihOS5LLJafV99XYxGddgwabKt8SmEyEQ%2FmRDlSoUA9gsNvKMDmhE8MC4L7OFtSYmPFmFlAmzm%2F9tfH0Oz8v6yFmxQ3SpOiY8eYTwjHew0%2BB9%2FD6B5ga4dLd%2FHQus0SnzaIrzWWgDb9P19MVqjw01dwFLpYYVYQymLgD1Kjj6J1umaHwLLqJfpy0%2FHIryqgg2mvetDKxXMnQMWEa9LxEpSqxZguS%2B%2BfA%2Bt9cZBi7ZxeqVMX376FqEnAtbyv7ISrTfspB%2FM82bq3r70BNMSYKV%2Bo4rQDiPzc8Csy1Fih%2BhVsE7o0cfQHnn%2FygJz6uNEJtaTSfy8ChYpnelDuxQ8HAIT1LOS8fwoCSq1FiVYcs%2FdaJ%2FgNhMJqrWKqfwoCSYtSTA08260U%2FBh47v4LDU%2F%2FgnmPOJDexX86ycwpp6yf80neB7M8o96DO2Wl2%2Bw%2FlLrh%2FlKYroW31qE9ht5EgzwRs3nR00wmgBTVq1EFtp2Ad0imdbkR0kwLQImTP8S2eg9B3QSKwkbHhPPxSUzAsjGe3P1luLrMmGklQpGjfIhKwU6C8llibBJUCaS4UKy6klkp0cX0CE9zcr8KAlei4Ahy36PLHXuBJqpYcJSmQBG3LIJWerQETS7qhCWlHowoMvfka2Va0Gjaus3MGUTp4NuWY8ja3%2FuB9q0IqydBt1eeQxZ%2B9MfQRNvnLAWT%2BiuIEuRvT9MBg3UlkQmbMmkUgB9cjsge8EbQIMLCmFPuQy6DPoGeVi9HqgED5EJazL5VAQ9Nm5CHjq0B6oKhZCUX4LrNyAfSycDhVBJZMKeTK4IoN26IPJRsAQoEhLhQ7kAmoV%2Bjbwspt0LniF8yKRMBa1%2B%2BSvkZVFfaFIkSngpvwha%2FQL56QNNqiX8%2FBs0mnMX8vPtBGiCWEf4iYmgzey7kZ8Rw6EJXonwo9SANn9GnuZCE84RnlqBJm3aIk8vFUKjxBjhKbMFaDHQhzy9%2BAI06pJEeJIS%2FGuwBn1M1WD%2BdXjNauSrdwk0Qq0kfHlUoFs7Evnq9TI0orqK8BVN1%2FIcvAn56vAKNCKhEDruz8NjkbdXOV4CKZJA1W8M8vbjT9CwMOGtDKjmjEbefpgCDRLqCB33p7kvipC3kc83UkOihLdohF5DfMjbiBf43UZTSPQq8vobyNsbudCgyzLhTT4PNK8hpmoZPkv4awU0y5G%2F1%2Fj90WG%2BDK9ATNX7mDDh71OgWYn83RHi9yRMkQY0I5G%2FOydDA4RPCX9RoMlD%2Fu6a0mCAMcJfHGh8yN%2BwqdAAMZPwJwFNB%2BRv5TRoQIs0wp%2FiiAB7TG%2B2Abor0L0GmiO5VdicuHsfaE7UfRIxJ80Rz8Kdnfss7L6NoShz8vvAWsLfOUe8kZ7o5DfSm1Pgm8gnTv4msqoIzXC%2FyrUZjWa434XdPxOoRZjiHjTD%2FTcGNm9Cg9y%2Fs9z%2FAymi1e4fqqZ4VPcfaQZnlQYGkacXP3H6X%2FrT2qIZ7jkR%2BAvy9L5jTyq5Z%2BUolBpHnNYc5PDTmubrsHtemOeJ9aJmcWI9tAV5%2BQ29Z4Kc%2Bj0TYHOQVwl5pVl07YD1h9EMt28MHOHUueihZtK5CArvRB4OTWkuvbNgYjGyF5wEGlQ4oXsbrF%2BK7O2fDBoIPPoHegQndLAc14w6WELot8jaX5pVD1Xo8iSy1WM8nzbcFMZbcf%2BLcR%2Fp7qBZayf0kYZly5GlzpOd3Mmcfy%2F9rl1AhwjTXvoXwaATDKc55Dp6mgP%2FeSLvZ4E%2B55wwTwSmr0Y2Djp6og3%2FmUrDhqbuTKWLYMqQ42i%2FkcNTdqpXeQ2Y4z82AO2Wl8txrpz5AkLRr38Q7TUiOydlJxueBfNCYzugnYKvOn62JkXpA3YmGPy8xPnTXanzhYP27d8PSvjPFzafH0Wov12VJC87ZSdcS2dVsEy%2FE8fRDgtznTFj3Tz%2FrT3QesOGO2bKv3mrVr%2BH1nrjjqFgiUilTGRr8%2FNEwHLTZ%2FisLR9vzgGLiOckYiWpVQuwQcmonmidZ3JDYBn1chohslXL79pVFWzh%2F2L5JrRG8fahYKlIWCHWUMoiYJtl%2F3wygOYFunabDBYTWmtdhJTlVy%2BAjfxPPP4YmpW3dTzYID0jTo%2BQEl88Ix1sFlqytAOacfe%2Bk1lgD29LxXiEMiFKZUIF%2By3L%2F6YYjSpu134w2EaouEKPsNH4rlwWgI0JEzcE0Qjfl19NAVsJFR6JGCF5LovAzrId2%2B8LoD6BBT8OGQy2E2rCUaJXebhGALZC9z%2FwUhC18%2F0wc1UWsBFJ1klEOymWvKgCe%2F7CW999xxdAusCI0R99PMgP7IiJczFJY3qtEiLw8tOckw88uKs40FR4xXuWzvzjVD%2BwJnqTlVUKaYpS5Ul6ReCsdOeOmVveKgq%2Bh%2F%2FvveCiu7Zvmz2rFDhRq2tqw7GoJJP%2FJ0vRWFmyplqF1NBv0KmTJz7fumX1d889%2B8yTzzz73Ldfbtm6bdS48RNygDcx3Xu1NqPMUxdLS7uWlhar85RlJK9600VIOf6c0mWDpj391NNtBg0uyfFDSlEF8T%2Ft3eFyqjwTwPGNiKq9eq%2BtqiCeoxZVEcRW4mK%2Bvc%2F5%2Bk7bBSDZOJPfFfwHWkEMG%2B%2BfXChwHMdxHMdxHMdxHMdxHMdxHIeV4yiR%2FyOUS6tHfBxP88Vse74N%2F7mdt7PF%2FHT8EFakbYg0XupvMZ%2Fddt%2F%2Ber27zebFX%2BXSfpQfD%2BMLsX7iMp4fc460%2BfgiqbSD1jSCGH1WXAV1v32OhOm0O1Yh9aUR0sNUYnVyekjBEH9eL%2B2mIY2gilmGdWXvhTKQNnpvkDYrBJgjNluJTchtIDSnBY3TNgLMUEGvbL4Qvhco3WkPbOS%2FNAEGjMay1bsEMjyCJsewXVo5HoFuH5P2b7OsJh9a0har1mn3tmkElXTzPlU%2FUd2nDfnTKH53b%2FTN%2FI7TZp2l7X3QZNPlO6X9jb1pJwUa5J8SuyQ%2Fc2vTFjl0zu%2F8vfrH2O8obdx52jaFjmmZ7HAdQQeOVw1pwxF0StNskd0GWtvsUIfsBB3SNt3m%2FgUtva1402jEfCXm%2BUBLjWkHBZ2gJ3zxHcG51JhWdnQENc%2BYk3O2vz%2F6CEJrBqYcyi9o6E172hJaMjJn876BRjYG0k7QiqFJr7tRo7SdgbSsgBaMzRoe%2BlCbfzWTlkILxqZdj%2FPaaWM0Y%2BtBUwbnrT8%2BoaZPY2kLBc2Ynfi%2FgVo2BtNO0JDRPSf6PtTgm0y7pNCI2KNJewWVqZnZNAH1md93J4HKEsNpb1Abw85P%2FQ%2Bo6GNoOs2H%2BgZo2gQqWqBpA6iNY%2Fe7EVRyXNm%2FMR%2FP%2FotjBRWokCFtK6AOrh1AA6ggkBxpG6hFnImzzLUFKNv2uOec5Q9Qw3kO7N%2BgmT7LjB81asuU1hNQXSyRhyyAULClxVDdHh%2FI4YEzIMzY0vZQWZQhlyyFX6V8aasIqnoinwP86oB8nlBRfkM%2Btxx%2BIaZWpNGf03zkCH4xYk0r7PiuTljALz6R0wQqya%2FI6ZrTHy78acS%2FCSd5hB8dmdNGdlyDCQfiGmz7dVhtkddWWZvWU0D72CGv3Qf84O%2BFP40Wl8irLOAHBXtaDLQDoq0fgnPk9gTaHrnt4Qcz5Bba8T2OcBPwLUGnWXAnmGbILfP5Lm%2BELLX3WSp9v3q0IC0GytcDuT1O8K2TBWlLq58kEJfhOfJbACVEfhN7z20IlDPy2xM3WIymQBkiv57i%2ByZM6ANlh%2FymAr6hpshvB5QVoqW3q%2BKK%2FO5AkchvmMM38iHyk0ApkV%2Ffg294feRXugPoDiCr0n0GtiPdVbid%2BwvfB4op8svcN5F2%2Bu67cDvTV34aM0F%2B4Ss%2FDzzYcW4JSwse%2Byav%2FETa4t9ERhakBS%2F9q5wFaRH%2F6kDaNbf3d2EPXuAyvLd30UQItCdyO9i7bOf5EquzYnvTgpdeH8iflvlAUz3kZf8KVcs%2FBJ%2F2rl1cQxWFvUvhR8xpBVThDfnvAu28SR16UMkEOS3sfdQxgGri0tp%2Fk0Lac39l6T%2FKLbd2AfLVg4rW9t7rPy24BtOiFXJZRda%2BTL%2F6A1Wp0N7BBHu2tFBBZUGJPGRs7QPfMrB9cBExnIV7pM1ZQA0nrvFA9qYlUEc%2B5R9QZddYrymdxn%2Bey5O9g%2BUSqEf0rB3SJ7YMaT0BNRUMEywLa9NkDHWpdzRtYO9413cFtaUXw6NyL76VA4abj%2BL%2BMjys%2BcvaEdePJTQhxmhSKGqkhWjSWEAj0cXagfWpybRdBA0lpbktExJrN5oo36ApNUFTJqpm2gJNGShozOuhGT3P2rSzBy1EfSMbF%2FVTqC01lBZBK%2FHK2q2zisxA2iqGlhKpf%2FO2pGHaXXuafOPfGZKMLJeMO0MSaXNoTz1LvRtYPhXftqlE2lpBB9SayOQ6fgDqqTXtk07jzKSPH00dpL60tbJ9h%2Bb2%2BzODWt7tSKM34tZhlUBrSaYn7Q06Ffc1bKXfj6EDhQ1ptOhcP5OI7EXQibTXedo5gs55gxK7VE68ztImstu0gQcaqGSH%2BOjqHF8S1WXapcO03ZsCPaLxA7tRhhF0Kg1L7MZjHIE24os%2B05X%2B%2FL6ErWm7pQCd0ndJdxKN93cfNPDf763T5CwFzVTcK%2BnOXxrLXqE0pRXbtmmxAv3EaUp3%2Ftg4PQlL0x7TRIAZeXIusYnyfMo1p50apyU5mCOCcIV1rcJA2J9mivqzvpZYXXldR8pQWlQ77Y8CBnk8GFYLlcNBnJtNmwwlVlH%2Bl%2BYBG69Yn7Py98Ksty48lrQemXY2kEZRfvAMr5l84P97yOwaPgNfWZq2NpZG86JgPhlP%2B9ldlo9S3rP%2BdDyZB5FnRdqygzTHcRzHcRzHcRzHcRzHcZz%2FAbyvLkVmYcs9AAAAAElFTkSuQmCC&link=https%3A%2F%2Fwww.together.ai" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
</div>
</div>
## DeepCoder Overview
DeepCoder-14B-Preview is a code reasoning LLM fine-tuned from DeepSeek-R1-Distilled-Qwen-14B using distributed reinforcement learning (RL) to scale up to long context lengths. The model achieves 60.6% Pass@1 accuracy on LiveCodeBench v5 (8/1/24-2/1/25), representing a 8% improvement over the base model (53%) and achieving similar performance to OpenAI's o3-mini with just 14B parameters.
<div style="margin: 0 auto;">
<img src="https://cdn-uploads.huggingface.co/production/uploads/654037be97949fd2304aab7f/r3-vzkItOCrMf1qldW0Mj.png" style="width: 100%;" />
</div>
## Data
Our training dataset consists of approximately 24K unique problem-tests pairs compiled from:
- Taco-Verified
- PrimeIntellect SYNTHETIC-1
- LiveCodeBench v5 (5/1/23-7/31/24)
## Training Recipe
Our training recipe relies on an improved version of GRPO (GRPO+) and iterative context lengthening, introduced in DeepScaleR.
### GRPO+
We enhance the original GRPO algorithm with insights from DAPO to enable more stable training:
- **Offline Difficulty Filtering:** DAPO employs online dynamic sampling, discarding both entirely correct and entirely incorrect samples on the fly. While this helps maintain a more stable effective batch size, it introduces significant runtime overhead due to rejection sampling. Instead, we perform offline difficulty filtering on a subset of coding problems to ensure the training dataset remains within a suitable difficulty range.
- **No Entropy Loss:** We observed that including an entropy loss term often led to instability, with entropy growing exponentially and ultimately collapsing training. To mitigate this, we eliminate the entropy loss entirely.
- **No KL Loss:** Eliminating KL loss prevents the LLM from staying within trust region of the original SFT model. This removal also obviates the need to compute log probabilities for the reference policy, thereby accelerating training.
- **Overlong Filtering** **(from DAPO):** To preserve long-context reasoning, we mask the loss for truncated sequences. This technique enables DeepCoder to generalize to 64K-context inference despite being trained with a 32K context.
- **Clip High (from DAPO):** By increasing the upper bound in GRPO/PPO’s surrogate loss, we encourage more exploration and more stable entropy.
### Iterative Context Lengthening
Our original `Deepscaler-1.5B-Preview` scaled long context training from 8K→16K→24K, achieving 33→38→43% on AIME respectively. Similarly, `Deepcoder-14B-Preview` is trained on 16K→32K, achieving 54→58% on LiveCodeBench (v5). `DeepCoder-14B-Preview` successfully generalizes to longer contexts when evaluated at 64K context, reaching 60.6%.
DeepCoder generalizes better to long contexts than the base distilled model, due to DAPO's overlong filtering. However, it's longer responses are often truncated when the max length is capped at 16K, which can lower its scores.
| **Model** | **16K** | **32K** | **64K** |
| --- | --- | --- | --- |
| **DeepCoder-14B-Preview** | 45.6 | 57.9 | 60.6 |
| **DeepSeek-R1-Distill-Qwen-14B** | 50.2 | 53.0 | 53.0 |
A more detailed description of the training recipe can be found in our [blog post](https://pretty-radio-b75.notion.site/DeepCoder-A-Fully-Open-Source-14B-Coder-at-O3-mini-Level-1cf81902c14680b3bee5eb349a512a51).
## Evaluation
We evaluate `Deepcoder-14B-Preview` on various coding benchmarks, including LiveCodeBench (LCBv5), Codeforces, and HumanEval+.
| **Model** | LCB (v5)(8/1/24-2/1/25) | Codeforces Rating | Codeforces Percentile | HumanEval+ |
| --- | --- | --- | --- | --- |
| **DeepCoder-14B-Preview (ours)** | ***60.6*** | ***1936*** | ***95.3*** | ***92.6*** |
| **DeepSeek-R1-Distill-Qwen-14B** | 53.0 | 1791 | 92.7 | 92.0 |
| **O1-2024-12-17 (Low)** | 59.5 | **1991** | **96.1** | 90.8 |
| **O3-Mini-2025-1-31 (Low)** | **60.9** | 1918 | 94.9 | 92.6 |
| **O1-Preview** | 42.7 | 1658 | 88.5 | 89 |
| **Deepseek-R1** | 62.8 | 1948 | 95.4 | 92.6 |
| **Llama-4-Behemoth** | 49.4 | - | - | - |
## Serving DeepCoder
Our model can be served using popular high-performance inference systems:
- vLLM
- Hugging Face Text Generation Inference (TGI)
- SGLang
- TensorRT-LLM
All these systems support the OpenAI Chat Completions API format.
### Usage Recommendations
Our usage recommendations are similar to those of R1 and R1 Distill series:
1. Avoid adding a system prompt; all instructions should be contained within the user prompt.
2. `temperature = 0.6`
3. `top_p = 0.95`
4. This model performs best with `max_tokens` set to at least `64000`
## License
This project is released under the MIT License, reflecting our commitment to open and accessible AI development.
We believe in democratizing AI technology by making our work freely available for anyone to use, modify, and build upon.
This permissive license ensures that researchers, developers, and enthusiasts worldwide can leverage and extend our work without restrictions, fostering innovation and collaboration in the AI community.
## Acknowledgement
- Our training experiments are powered by our heavily modified fork of [Verl](https://github.com/agentica-project/verl), an open-source post-training library.
- Our model is trained on top of [`DeepSeek-R1-Distill-Qwen-14B`](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B).
- Our work is done as part of [Berkeley Sky Computing Lab](https://skycomputing.berkeley.edu/) and [Berkeley AI Research](https://bair.berkeley.edu/).
## Citation
```bibtex
@misc{deepcoder2025,
title={DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level},
author={Michael Luo, Sijun Tan, Roy Huang, Ameen Patel, Alpay Ariyak, Qingyang Wu, Xiaoxiang Shi, Rachel Xin, Colin Cai, Maurice Weber, Ce Zhang, Li Erran Li, Raluca Ada Popa, Ion Stoica},
howpublished={\url{https://pretty-radio-b75.notion.site/DeepCoder-A-Fully-Open-Source-14B-Coder-at-O3-mini-Level-1cf81902c14680b3bee5eb349a512a51}},
note={Notion Blog},
year={2025}
}
``` |
Mungert/all-MiniLM-L6-v2-GGUF | Mungert | 2025-06-15T19:36:56Z | 2,709 | 4 | sentence-transformers | [
"sentence-transformers",
"gguf",
"feature-extraction",
"sentence-similarity",
"transformers",
"en",
"dataset:s2orc",
"dataset:flax-sentence-embeddings/stackexchange_xml",
"dataset:ms_marco",
"dataset:gooaq",
"dataset:yahoo_answers_topics",
"dataset:code_search_net",
"dataset:search_qa",
"dataset:eli5",
"dataset:snli",
"dataset:multi_nli",
"dataset:wikihow",
"dataset:natural_questions",
"dataset:trivia_qa",
"dataset:embedding-data/sentence-compression",
"dataset:embedding-data/flickr30k-captions",
"dataset:embedding-data/altlex",
"dataset:embedding-data/simple-wiki",
"dataset:embedding-data/QQP",
"dataset:embedding-data/SPECTER",
"dataset:embedding-data/PAQ_pairs",
"dataset:embedding-data/WikiAnswers",
"arxiv:1904.06472",
"arxiv:2102.07033",
"arxiv:2104.08727",
"arxiv:1704.05179",
"arxiv:1810.09305",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us",
"imatrix"
] | sentence-similarity | 2025-03-23T03:05:06Z | ---
language: en
license: apache-2.0
library_name: sentence-transformers
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
datasets:
- s2orc
- flax-sentence-embeddings/stackexchange_xml
- ms_marco
- gooaq
- yahoo_answers_topics
- code_search_net
- search_qa
- eli5
- snli
- multi_nli
- wikihow
- natural_questions
- trivia_qa
- embedding-data/sentence-compression
- embedding-data/flickr30k-captions
- embedding-data/altlex
- embedding-data/simple-wiki
- embedding-data/QQP
- embedding-data/SPECTER
- embedding-data/PAQ_pairs
- embedding-data/WikiAnswers
pipeline_tag: sentence-similarity
---
# <span style="color: #7FFF7F;">all-MiniLM-L6-v2 GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `all-MiniLM-L6-v2-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `all-MiniLM-L6-v2-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `all-MiniLM-L6-v2-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `all-MiniLM-L6-v2-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `all-MiniLM-L6-v2-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `all-MiniLM-L6-v2-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `all-MiniLM-L6-v2-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `all-MiniLM-L6-v2-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `all-MiniLM-L6-v2-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `all-MiniLM-L6-v2-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `all-MiniLM-L6-v2-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download/?utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme) the Quantum Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
### Final word
I fund the servers to create the models files, run the Quantum Network Monitor Service and Pay for Inference from Novita and OpenAI all from my own pocket. All of the code for creating the models and the work I have done with Quantum Network Monitor is [open source](https://github.com/Mungert69). Feel free to use what you find useful. Please support my work and consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) .
This will help me pay for the services and increase the token limits for everyone.
Thank you :)
# all-MiniLM-L6-v2
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search.
## Usage (Sentence-Transformers)
Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
```
pip install -U sentence-transformers
```
Then you can use the model like this:
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
embeddings = model.encode(sentences)
print(embeddings)
```
## Usage (HuggingFace Transformers)
Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
```python
from transformers import AutoTokenizer, AutoModel
import torch
import torch.nn.functional as F
#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
token_embeddings = model_output[0] #First element of model_output contains all token embeddings
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
model = AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
# Compute token embeddings
with torch.no_grad():
model_output = model(**encoded_input)
# Perform pooling
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
# Normalize embeddings
sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)
print("Sentence embeddings:")
print(sentence_embeddings)
```
------
## Background
The project aims to train sentence embedding models on very large sentence level datasets using a self-supervised
contrastive learning objective. We used the pretrained [`nreimers/MiniLM-L6-H384-uncased`](https://huggingface.co/nreimers/MiniLM-L6-H384-uncased) model and fine-tuned in on a
1B sentence pairs dataset. We use a contrastive learning objective: given a sentence from the pair, the model should predict which out of a set of randomly sampled other sentences, was actually paired with it in our dataset.
We developed this model during the
[Community week using JAX/Flax for NLP & CV](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104),
organized by Hugging Face. We developed this model as part of the project:
[Train the Best Sentence Embedding Model Ever with 1B Training Pairs](https://discuss.huggingface.co/t/train-the-best-sentence-embedding-model-ever-with-1b-training-pairs/7354). We benefited from efficient hardware infrastructure to run the project: 7 TPUs v3-8, as well as intervention from Googles Flax, JAX, and Cloud team member about efficient deep learning frameworks.
## Intended uses
Our model is intended to be used as a sentence and short paragraph encoder. Given an input text, it outputs a vector which captures
the semantic information. The sentence vector may be used for information retrieval, clustering or sentence similarity tasks.
By default, input text longer than 256 word pieces is truncated.
## Training procedure
### Pre-training
We use the pretrained [`nreimers/MiniLM-L6-H384-uncased`](https://huggingface.co/nreimers/MiniLM-L6-H384-uncased) model. Please refer to the model card for more detailed information about the pre-training procedure.
### Fine-tuning
We fine-tune the model using a contrastive objective. Formally, we compute the cosine similarity from each possible sentence pairs from the batch.
We then apply the cross entropy loss by comparing with true pairs.
#### Hyper parameters
We trained our model on a TPU v3-8. We train the model during 100k steps using a batch size of 1024 (128 per TPU core).
We use a learning rate warm up of 500. The sequence length was limited to 128 tokens. We used the AdamW optimizer with
a 2e-5 learning rate. The full training script is accessible in this current repository: `train_script.py`.
#### Training data
We use the concatenation from multiple datasets to fine-tune our model. The total number of sentence pairs is above 1 billion sentences.
We sampled each dataset given a weighted probability which configuration is detailed in the `data_config.json` file.
| Dataset | Paper | Number of training tuples |
|--------------------------------------------------------|:----------------------------------------:|:--------------------------:|
| [Reddit comments (2015-2018)](https://github.com/PolyAI-LDN/conversational-datasets/tree/master/reddit) | [paper](https://arxiv.org/abs/1904.06472) | 726,484,430 |
| [S2ORC](https://github.com/allenai/s2orc) Citation pairs (Abstracts) | [paper](https://aclanthology.org/2020.acl-main.447/) | 116,288,806 |
| [WikiAnswers](https://github.com/afader/oqa#wikianswers-corpus) Duplicate question pairs | [paper](https://doi.org/10.1145/2623330.2623677) | 77,427,422 |
| [PAQ](https://github.com/facebookresearch/PAQ) (Question, Answer) pairs | [paper](https://arxiv.org/abs/2102.07033) | 64,371,441 |
| [S2ORC](https://github.com/allenai/s2orc) Citation pairs (Titles) | [paper](https://aclanthology.org/2020.acl-main.447/) | 52,603,982 |
| [S2ORC](https://github.com/allenai/s2orc) (Title, Abstract) | [paper](https://aclanthology.org/2020.acl-main.447/) | 41,769,185 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) (Title, Body) pairs | - | 25,316,456 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) (Title+Body, Answer) pairs | - | 21,396,559 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) (Title, Answer) pairs | - | 21,396,559 |
| [MS MARCO](https://microsoft.github.io/msmarco/) triplets | [paper](https://doi.org/10.1145/3404835.3462804) | 9,144,553 |
| [GOOAQ: Open Question Answering with Diverse Answer Types](https://github.com/allenai/gooaq) | [paper](https://arxiv.org/pdf/2104.08727.pdf) | 3,012,496 |
| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Title, Answer) | [paper](https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html) | 1,198,260 |
| [Code Search](https://huggingface.co/datasets/code_search_net) | - | 1,151,414 |
| [COCO](https://cocodataset.org/#home) Image captions | [paper](https://link.springer.com/chapter/10.1007%2F978-3-319-10602-1_48) | 828,395|
| [SPECTER](https://github.com/allenai/specter) citation triplets | [paper](https://doi.org/10.18653/v1/2020.acl-main.207) | 684,100 |
| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Question, Answer) | [paper](https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html) | 681,164 |
| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Title, Question) | [paper](https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html) | 659,896 |
| [SearchQA](https://huggingface.co/datasets/search_qa) | [paper](https://arxiv.org/abs/1704.05179) | 582,261 |
| [Eli5](https://huggingface.co/datasets/eli5) | [paper](https://doi.org/10.18653/v1/p19-1346) | 325,475 |
| [Flickr 30k](https://shannon.cs.illinois.edu/DenotationGraph/) | [paper](https://transacl.org/ojs/index.php/tacl/article/view/229/33) | 317,695 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) Duplicate questions (titles) | | 304,525 |
| AllNLI ([SNLI](https://nlp.stanford.edu/projects/snli/) and [MultiNLI](https://cims.nyu.edu/~sbowman/multinli/) | [paper SNLI](https://doi.org/10.18653/v1/d15-1075), [paper MultiNLI](https://doi.org/10.18653/v1/n18-1101) | 277,230 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) Duplicate questions (bodies) | | 250,519 |
| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) Duplicate questions (titles+bodies) | | 250,460 |
| [Sentence Compression](https://github.com/google-research-datasets/sentence-compression) | [paper](https://www.aclweb.org/anthology/D13-1155/) | 180,000 |
| [Wikihow](https://github.com/pvl/wikihow_pairs_dataset) | [paper](https://arxiv.org/abs/1810.09305) | 128,542 |
| [Altlex](https://github.com/chridey/altlex/) | [paper](https://aclanthology.org/P16-1135.pdf) | 112,696 |
| [Quora Question Triplets](https://quoradata.quora.com/First-Quora-Dataset-Release-Question-Pairs) | - | 103,663 |
| [Simple Wikipedia](https://cs.pomona.edu/~dkauchak/simplification/) | [paper](https://www.aclweb.org/anthology/P11-2117/) | 102,225 |
| [Natural Questions (NQ)](https://ai.google.com/research/NaturalQuestions) | [paper](https://transacl.org/ojs/index.php/tacl/article/view/1455) | 100,231 |
| [SQuAD2.0](https://rajpurkar.github.io/SQuAD-explorer/) | [paper](https://aclanthology.org/P18-2124.pdf) | 87,599 |
| [TriviaQA](https://huggingface.co/datasets/trivia_qa) | - | 73,346 |
| **Total** | | **1,170,060,424** | |
Mungert/Meta-Llama-3-8B-Instruct-GGUF | Mungert | 2025-06-15T19:36:50Z | 3,003 | 4 | null | [
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"license:llama3",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-03-17T05:54:18Z | ---
language:
- en
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3
new_version: meta-llama/Llama-3.1-8B-Instruct
extra_gated_prompt: >-
### META LLAMA 3 COMMUNITY LICENSE AGREEMENT
Meta Llama 3 Version Release Date: April 18, 2024
"Agreement" means the terms and conditions for use, reproduction, distribution and modification of the
Llama Materials set forth herein.
"Documentation" means the specifications, manuals and documentation accompanying Meta Llama 3
distributed by Meta at https://llama.meta.com/get-started/.
"Licensee" or "you" means you, or your employer or any other person or entity (if you are entering into
this Agreement on such person or entity’s behalf), of the age required under applicable laws, rules or
regulations to provide legal consent and that has legal authority to bind your employer or such other
person or entity if you are entering in this Agreement on their behalf.
"Meta Llama 3" means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://llama.meta.com/llama-downloads.
"Llama Materials" means, collectively, Meta’s proprietary Meta Llama 3 and Documentation (and any
portion thereof) made available under this Agreement.
"Meta" or "we" means Meta Platforms Ireland Limited (if you are located in or, if you are an entity, your
principal place of business is in the EEA or Switzerland) and Meta Platforms, Inc. (if you are located
outside of the EEA or Switzerland).
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide, non-transferable and royalty-free
limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama
Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the
Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works
thereof), or a product or service that uses any of them, including another AI model, you shall (A) provide
a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Meta
Llama 3” on a related website, user interface, blogpost, about page, or product documentation. If you
use the Llama Materials to create, train, fine tune, or otherwise improve an AI model, which is
distributed or made available, you shall also include “Llama 3” at the beginning of any such AI model
name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the following
attribution notice within a “Notice” text file distributed as a part of such copies: “Meta Llama 3 is
licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights
Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for the Llama
Materials (available at https://llama.meta.com/llama3/use-policy), which is hereby incorporated by
reference into this Agreement.
v. You will not use the Llama Materials or any output or results of the Llama Materials to
improve any other large language model (excluding Meta Llama 3 or derivative works thereof).
2. Additional Commercial Terms. If, on the Meta Llama 3 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700
million monthly active users in the preceding calendar month, you must request a license from Meta,
which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the
rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY
OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF
ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED,
INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT,
MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR
DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND
ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING
OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL,
INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN IF META OR ITS AFFILIATES HAVE BEEN ADVISED
OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama
Materials, neither Meta nor Licensee may use any name or mark owned by or associated with the other
or any of its affiliates, except as required for reasonable and customary use in describing and
redistributing the Llama Materials or as set forth in this Section 5(a). Meta hereby grants you a license to
use “Llama 3” (the “Mark”) solely as required to comply with the last sentence of Section 1.b.i. You will
comply with Meta’s brand guidelines (currently accessible at
https://about.meta.com/brand/resources/meta/company-brand/ ). All goodwill arising out of your use
of the Mark will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with
respect to any derivative works and modifications of the Llama Materials that are made by you, as
between you and Meta, you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Meta Llama 3 outputs or
results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other
rights owned or licensable by you, then any licenses granted to you under this Agreement shall
terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold
harmless Meta from and against any claim by any third party arising out of or related to your use or
distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this
Agreement or access to the Llama Materials and will continue in full force and effect until terminated in
accordance with the terms and conditions herein. Meta may terminate this Agreement if you are in
breach of any term or condition of this Agreement. Upon termination of this Agreement, you shall delete
and cease use of the Llama Materials. Sections 3, 4 and 7 shall survive the termination of this
Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of
the State of California without regard to choice of law principles, and the UN Convention on Contracts
for the International Sale of Goods does not apply to this Agreement. The courts of California shall have
exclusive jurisdiction of any dispute arising out of this Agreement.
### Meta Llama 3 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Meta Llama 3. If you
access or use Meta Llama 3, you agree to this Acceptable Use Policy (“Policy”). The most recent copy of
this policy can be found at [https://llama.meta.com/llama3/use-policy](https://llama.meta.com/llama3/use-policy)
#### Prohibited Uses
We want everyone to use Meta Llama 3 safely and responsibly. You agree you will not use, or allow
others to use, Meta Llama 3 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
2. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
3. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
4. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
5. Collect, process, disclose, generate, or infer health, demographic, or other sensitive personal or private information about individuals without rights and consents required by applicable laws
6. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
7. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Meta Llama 3 related to the following:
1. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State
2. Guns and illegal weapons (including weapon development)
3. Illegal drugs and regulated/controlled substances
4. Operation of critical infrastructure, transportation technologies, or heavy machinery
5. Self-harm or harm to others, including suicide, cutting, and eating disorders
6. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Meta Llama 3 related to the following:
1. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
2. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
3. Generating, promoting, or further distributing spam
4. Impersonating another individual without consent, authorization, or legal right
5. Representing that the use of Meta Llama 3 or outputs are human-generated
6. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation
of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama3](https://github.com/meta-llama/llama3)
* Reporting risky content generated by the model:
developers.facebook.com/llama_output_feedback
* Reporting bugs and security concerns: facebook.com/whitehat/info
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Meta Llama 3: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: The information you provide will be collected, stored, processed and shared in accordance with the [Meta Privacy Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
widget:
- example_title: Hello
messages:
- role: user
content: Hey my name is Julien! How are you?
- example_title: Winter holidays
messages:
- role: system
content: You are a helpful and honest assistant. Please, respond concisely and truthfully.
- role: user
content: Can you recommend a good destination for Winter holidays?
- example_title: Programming assistant
messages:
- role: system
content: You are a helpful and honest code and programming assistant. Please, respond concisely and truthfully.
- role: user
content: Write a function that computes the nth fibonacci number.
inference:
parameters:
max_new_tokens: 300
stop:
- <|end_of_text|>
- <|eot_id|>
---
# <span style="color: #7FFF7F;">Meta-Llama-3-8B-Instruct GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`f5cd27b7`](https://github.com/ggerganov/llama.cpp/commit/f5cd27b71da3ac375a04a41643d14fc779a8057b).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Meta-Llama-3-8B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Meta-Llama-3-8B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Meta-Llama-3-8B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Meta-Llama-3-8B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Meta-Llama-3-8B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Meta-Llama-3-8B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Meta-Llama-3-8B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Meta-Llama-3-8B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Meta-Llama-3-8B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Meta-Llama-3-8B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Meta-Llama-3-8B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
## Model Details
Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.
**Model developers** Meta
**Variations** Llama 3 comes in two sizes — 8B and 70B parameters — in pre-trained and instruction tuned variants.
**Input** Models input text only.
**Output** Models generate text and code only.
**Model Architecture** Llama 3 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
<table>
<tr>
<td>
</td>
<td><strong>Training Data</strong>
</td>
<td><strong>Params</strong>
</td>
<td><strong>Context length</strong>
</td>
<td><strong>GQA</strong>
</td>
<td><strong>Token count</strong>
</td>
<td><strong>Knowledge cutoff</strong>
</td>
</tr>
<tr>
<td rowspan="2" >Llama 3
</td>
<td rowspan="2" >A new mix of publicly available online data.
</td>
<td>8B
</td>
<td>8k
</td>
<td>Yes
</td>
<td rowspan="2" >15T+
</td>
<td>March, 2023
</td>
</tr>
<tr>
<td>70B
</td>
<td>8k
</td>
<td>Yes
</td>
<td>December, 2023
</td>
</tr>
</table>
**Llama 3 family of models**. Token counts refer to pretraining data only. Both the 8 and 70B versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date** April 18, 2024.
**Status** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.
**License** A custom commercial license is available at: [https://llama.meta.com/llama3/license](https://llama.meta.com/llama3/license)
Where to send questions or comments about the model Instructions on how to provide feedback or comments on the model can be found in the model [README](https://github.com/meta-llama/llama3). For more technical information about generation parameters and recipes for how to use Llama 3 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases** Llama 3 is intended for commercial and research use in English. Instruction tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.
**Out-of-scope** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3 Community License. Use in languages other than English**.
**Note: Developers may fine-tune Llama 3 models for languages beyond English provided they comply with the Llama 3 Community License and the Acceptable Use Policy.
## How to use
This repository contains two versions of Meta-Llama-3-8B-Instruct, for use with transformers and with the original `llama3` codebase.
### Use with transformers
You can run conversational inference using the Transformers pipeline abstraction, or by leveraging the Auto classes with the `generate()` function. Let's see examples of both.
#### Transformers pipeline
```python
import transformers
import torch
model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
terminators = [
pipeline.tokenizer.eos_token_id,
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = pipeline(
messages,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9,
)
print(outputs[0]["generated_text"][-1])
```
#### Transformers AutoModelForCausalLM
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
input_ids = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_tensors="pt"
).to(model.device)
terminators = [
tokenizer.eos_token_id,
tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = model.generate(
input_ids,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```
### Use with `llama3`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama3)
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Meta-Llama-3-8B-Instruct --include "original/*" --local-dir Meta-Llama-3-8B-Instruct
```
For Hugging Face support, we recommend using transformers or TGI, but a similar command works.
## Hardware and Software
**Training Factors** We used custom training libraries, Meta's Research SuperCluster, and production clusters for pretraining. Fine-tuning, annotation, and evaluation were also performed on third-party cloud compute.
**Carbon Footprint Pretraining utilized a cumulative** 7.7M GPU hours of computation on hardware of type H100-80GB (TDP of 700W). Estimated total emissions were 2290 tCO2eq, 100% of which were offset by Meta’s sustainability program.
<table>
<tr>
<td>
</td>
<td><strong>Time (GPU hours)</strong>
</td>
<td><strong>Power Consumption (W)</strong>
</td>
<td><strong>Carbon Emitted(tCO2eq)</strong>
</td>
</tr>
<tr>
<td>Llama 3 8B
</td>
<td>1.3M
</td>
<td>700
</td>
<td>390
</td>
</tr>
<tr>
<td>Llama 3 70B
</td>
<td>6.4M
</td>
<td>700
</td>
<td>1900
</td>
</tr>
<tr>
<td>Total
</td>
<td>7.7M
</td>
<td>
</td>
<td>2290
</td>
</tr>
</table>
**CO2 emissions during pre-training**. Time: total GPU time required for training each model. Power Consumption: peak power capacity per GPU device for the GPUs used adjusted for power usage efficiency. 100% of the emissions are directly offset by Meta's sustainability program, and because we are openly releasing these models, the pretraining costs do not need to be incurred by others.
## Training Data
**Overview** Llama 3 was pretrained on over 15 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over 10M human-annotated examples. Neither the pretraining nor the fine-tuning datasets include Meta user data.
**Data Freshness** The pretraining data has a cutoff of March 2023 for the 8B and December 2023 for the 70B models respectively.
## Benchmarks
In this section, we report the results for Llama 3 models on standard automatic benchmarks. For all the evaluations, we use our internal evaluations library. For details on the methodology see [here](https://github.com/meta-llama/llama3/blob/main/eval_methodology.md).
### Base pretrained models
<table>
<tr>
<td><strong>Category</strong>
</td>
<td><strong>Benchmark</strong>
</td>
<td><strong>Llama 3 8B</strong>
</td>
<td><strong>Llama2 7B</strong>
</td>
<td><strong>Llama2 13B</strong>
</td>
<td><strong>Llama 3 70B</strong>
</td>
<td><strong>Llama2 70B</strong>
</td>
</tr>
<tr>
<td rowspan="6" >General
</td>
<td>MMLU (5-shot)
</td>
<td>66.6
</td>
<td>45.7
</td>
<td>53.8
</td>
<td>79.5
</td>
<td>69.7
</td>
</tr>
<tr>
<td>AGIEval English (3-5 shot)
</td>
<td>45.9
</td>
<td>28.8
</td>
<td>38.7
</td>
<td>63.0
</td>
<td>54.8
</td>
</tr>
<tr>
<td>CommonSenseQA (7-shot)
</td>
<td>72.6
</td>
<td>57.6
</td>
<td>67.6
</td>
<td>83.8
</td>
<td>78.7
</td>
</tr>
<tr>
<td>Winogrande (5-shot)
</td>
<td>76.1
</td>
<td>73.3
</td>
<td>75.4
</td>
<td>83.1
</td>
<td>81.8
</td>
</tr>
<tr>
<td>BIG-Bench Hard (3-shot, CoT)
</td>
<td>61.1
</td>
<td>38.1
</td>
<td>47.0
</td>
<td>81.3
</td>
<td>65.7
</td>
</tr>
<tr>
<td>ARC-Challenge (25-shot)
</td>
<td>78.6
</td>
<td>53.7
</td>
<td>67.6
</td>
<td>93.0
</td>
<td>85.3
</td>
</tr>
<tr>
<td>Knowledge reasoning
</td>
<td>TriviaQA-Wiki (5-shot)
</td>
<td>78.5
</td>
<td>72.1
</td>
<td>79.6
</td>
<td>89.7
</td>
<td>87.5
</td>
</tr>
<tr>
<td rowspan="4" >Reading comprehension
</td>
<td>SQuAD (1-shot)
</td>
<td>76.4
</td>
<td>72.2
</td>
<td>72.1
</td>
<td>85.6
</td>
<td>82.6
</td>
</tr>
<tr>
<td>QuAC (1-shot, F1)
</td>
<td>44.4
</td>
<td>39.6
</td>
<td>44.9
</td>
<td>51.1
</td>
<td>49.4
</td>
</tr>
<tr>
<td>BoolQ (0-shot)
</td>
<td>75.7
</td>
<td>65.5
</td>
<td>66.9
</td>
<td>79.0
</td>
<td>73.1
</td>
</tr>
<tr>
<td>DROP (3-shot, F1)
</td>
<td>58.4
</td>
<td>37.9
</td>
<td>49.8
</td>
<td>79.7
</td>
<td>70.2
</td>
</tr>
</table>
### Instruction tuned models
<table>
<tr>
<td><strong>Benchmark</strong>
</td>
<td><strong>Llama 3 8B</strong>
</td>
<td><strong>Llama 2 7B</strong>
</td>
<td><strong>Llama 2 13B</strong>
</td>
<td><strong>Llama 3 70B</strong>
</td>
<td><strong>Llama 2 70B</strong>
</td>
</tr>
<tr>
<td>MMLU (5-shot)
</td>
<td>68.4
</td>
<td>34.1
</td>
<td>47.8
</td>
<td>82.0
</td>
<td>52.9
</td>
</tr>
<tr>
<td>GPQA (0-shot)
</td>
<td>34.2
</td>
<td>21.7
</td>
<td>22.3
</td>
<td>39.5
</td>
<td>21.0
</td>
</tr>
<tr>
<td>HumanEval (0-shot)
</td>
<td>62.2
</td>
<td>7.9
</td>
<td>14.0
</td>
<td>81.7
</td>
<td>25.6
</td>
</tr>
<tr>
<td>GSM-8K (8-shot, CoT)
</td>
<td>79.6
</td>
<td>25.7
</td>
<td>77.4
</td>
<td>93.0
</td>
<td>57.5
</td>
</tr>
<tr>
<td>MATH (4-shot, CoT)
</td>
<td>30.0
</td>
<td>3.8
</td>
<td>6.7
</td>
<td>50.4
</td>
<td>11.6
</td>
</tr>
</table>
### Responsibility & Safety
We believe that an open approach to AI leads to better, safer products, faster innovation, and a bigger overall market. We are committed to Responsible AI development and took a series of steps to limit misuse and harm and support the open source community.
Foundation models are widely capable technologies that are built to be used for a diverse range of applications. They are not designed to meet every developer preference on safety levels for all use cases, out-of-the-box, as those by their nature will differ across different applications.
Rather, responsible LLM-application deployment is achieved by implementing a series of safety best practices throughout the development of such applications, from the model pre-training, fine-tuning and the deployment of systems composed of safeguards to tailor the safety needs specifically to the use case and audience.
As part of the Llama 3 release, we updated our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/) to outline the steps and best practices for developers to implement model and system level safety for their application. We also provide a set of resources including [Meta Llama Guard 2](https://llama.meta.com/purple-llama/) and [Code Shield](https://llama.meta.com/purple-llama/) safeguards. These tools have proven to drastically reduce residual risks of LLM Systems, while maintaining a high level of helpfulness. We encourage developers to tune and deploy these safeguards according to their needs and we provide a [reference implementation](https://github.com/meta-llama/llama-recipes/tree/main/recipes/responsible_ai) to get you started.
#### Llama 3-Instruct
As outlined in the Responsible Use Guide, some trade-off between model helpfulness and model alignment is likely unavoidable. Developers should exercise discretion about how to weigh the benefits of alignment and helpfulness for their specific use case and audience. Developers should be mindful of residual risks when using Llama models and leverage additional safety tools as needed to reach the right safety bar for their use case.
<span style="text-decoration:underline;">Safety</span>
For our instruction tuned model, we conducted extensive red teaming exercises, performed adversarial evaluations and implemented safety mitigations techniques to lower residual risks. As with any Large Language Model, residual risks will likely remain and we recommend that developers assess these risks in the context of their use case. In parallel, we are working with the community to make AI safety benchmark standards transparent, rigorous and interpretable.
<span style="text-decoration:underline;">Refusals</span>
In addition to residual risks, we put a great emphasis on model refusals to benign prompts. Over-refusing not only can impact the user experience but could even be harmful in certain contexts as well. We’ve heard the feedback from the developer community and improved our fine tuning to ensure that Llama 3 is significantly less likely to falsely refuse to answer prompts than Llama 2.
We built internal benchmarks and developed mitigations to limit false refusals making Llama 3 our most helpful model to date.
#### Responsible release
In addition to responsible use considerations outlined above, we followed a rigorous process that requires us to take extra measures against misuse and critical risks before we make our release decision.
Misuse
If you access or use Llama 3, you agree to the Acceptable Use Policy. The most recent copy of this policy can be found at [https://llama.meta.com/llama3/use-policy/](https://llama.meta.com/llama3/use-policy/).
#### Critical risks
<span style="text-decoration:underline;">CBRNE</span> (Chemical, Biological, Radiological, Nuclear, and high yield Explosives)
We have conducted a two fold assessment of the safety of the model in this area:
* Iterative testing during model training to assess the safety of responses related to CBRNE threats and other adversarial risks.
* Involving external CBRNE experts to conduct an uplift test assessing the ability of the model to accurately provide expert knowledge and reduce barriers to potential CBRNE misuse, by reference to what can be achieved using web search (without the model).
### <span style="text-decoration:underline;">Cyber Security </span>
We have evaluated Llama 3 with CyberSecEval, Meta’s cybersecurity safety eval suite, measuring Llama 3’s propensity to suggest insecure code when used as a coding assistant, and Llama 3’s propensity to comply with requests to help carry out cyber attacks, where attacks are defined by the industry standard MITRE ATT&CK cyber attack ontology. On our insecure coding and cyber attacker helpfulness tests, Llama 3 behaved in the same range or safer than models of [equivalent coding capability](https://huggingface.co/spaces/facebook/CyberSecEval).
### <span style="text-decoration:underline;">Child Safety</span>
Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
### Community
Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership in AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
The core values of Llama 3 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
But Llama 3 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has been in English, and has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3 models, developers should perform safety testing and tuning tailored to their specific applications of the model. As outlined in the Responsible Use Guide, we recommend incorporating [Purple Llama](https://github.com/facebookresearch/PurpleLlama) solutions into your workflows and specifically [Llama Guard](https://ai.meta.com/research/publications/llama-guard-llm-based-input-output-safeguard-for-human-ai-conversations/) which provides a base model to filter input and output prompts to layer system-level safety on top of model-level safety.
Please see the Responsible Use Guide available at [http://llama.meta.com/responsible-use-guide](http://llama.meta.com/responsible-use-guide)
## Citation instructions
@article{llama3modelcard,
title={Llama 3 Model Card},
author={AI@Meta},
year={2024},
url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
}
## Contributors
Aaditya Singh; Aaron Grattafiori; Abhimanyu Dubey; Abhinav Jauhri; Abhinav Pandey; Abhishek Kadian; Adam Kelsey; Adi Gangidi; Ahmad Al-Dahle; Ahuva Goldstand; Aiesha Letman; Ajay Menon; Akhil Mathur; Alan Schelten; Alex Vaughan; Amy Yang; Andrei Lupu; Andres Alvarado; Andrew Gallagher; Andrew Gu; Andrew Ho; Andrew Poulton; Andrew Ryan; Angela Fan; Ankit Ramchandani; Anthony Hartshorn; Archi Mitra; Archie Sravankumar; Artem Korenev; Arun Rao; Ashley Gabriel; Ashwin Bharambe; Assaf Eisenman; Aston Zhang; Aurelien Rodriguez; Austen Gregerson; Ava Spataru; Baptiste Roziere; Ben Maurer; Benjamin Leonhardi; Bernie Huang; Bhargavi Paranjape; Bing Liu; Binh Tang; Bobbie Chern; Brani Stojkovic; Brian Fuller; Catalina Mejia Arenas; Chao Zhou; Charlotte Caucheteux; Chaya Nayak; Ching-Hsiang Chu; Chloe Bi; Chris Cai; Chris Cox; Chris Marra; Chris McConnell; Christian Keller; Christoph Feichtenhofer; Christophe Touret; Chunyang Wu; Corinne Wong; Cristian Canton Ferrer; Damien Allonsius; Daniel Kreymer; Daniel Haziza; Daniel Li; Danielle Pintz; Danny Livshits; Danny Wyatt; David Adkins; David Esiobu; David Xu; Davide Testuggine; Delia David; Devi Parikh; Dhruv Choudhary; Dhruv Mahajan; Diana Liskovich; Diego Garcia-Olano; Diego Perino; Dieuwke Hupkes; Dingkang Wang; Dustin Holland; Egor Lakomkin; Elina Lobanova; Xiaoqing Ellen Tan; Emily Dinan; Eric Smith; Erik Brinkman; Esteban Arcaute; Filip Radenovic; Firat Ozgenel; Francesco Caggioni; Frank Seide; Frank Zhang; Gabriel Synnaeve; Gabriella Schwarz; Gabrielle Lee; Gada Badeer; Georgia Anderson; Graeme Nail; Gregoire Mialon; Guan Pang; Guillem Cucurell; Hailey Nguyen; Hannah Korevaar; Hannah Wang; Haroun Habeeb; Harrison Rudolph; Henry Aspegren; Hu Xu; Hugo Touvron; Iga Kozlowska; Igor Molybog; Igor Tufanov; Iliyan Zarov; Imanol Arrieta Ibarra; Irina-Elena Veliche; Isabel Kloumann; Ishan Misra; Ivan Evtimov; Jacob Xu; Jade Copet; Jake Weissman; Jan Geffert; Jana Vranes; Japhet Asher; Jason Park; Jay Mahadeokar; Jean-Baptiste Gaya; Jeet Shah; Jelmer van der Linde; Jennifer Chan; Jenny Hong; Jenya Lee; Jeremy Fu; Jeremy Teboul; Jianfeng Chi; Jianyu Huang; Jie Wang; Jiecao Yu; Joanna Bitton; Joe Spisak; Joelle Pineau; Jon Carvill; Jongsoo Park; Joseph Rocca; Joshua Johnstun; Junteng Jia; Kalyan Vasuden Alwala; Kam Hou U; Kate Plawiak; Kartikeya Upasani; Kaushik Veeraraghavan; Ke Li; Kenneth Heafield; Kevin Stone; Khalid El-Arini; Krithika Iyer; Kshitiz Malik; Kuenley Chiu; Kunal Bhalla; Kyle Huang; Lakshya Garg; Lauren Rantala-Yeary; Laurens van der Maaten; Lawrence Chen; Leandro Silva; Lee Bell; Lei Zhang; Liang Tan; Louis Martin; Lovish Madaan; Luca Wehrstedt; Lukas Blecher; Luke de Oliveira; Madeline Muzzi; Madian Khabsa; Manav Avlani; Mannat Singh; Manohar Paluri; Mark Zuckerberg; Marcin Kardas; Martynas Mankus; Mathew Oldham; Mathieu Rita; Matthew Lennie; Maya Pavlova; Meghan Keneally; Melanie Kambadur; Mihir Patel; Mikayel Samvelyan; Mike Clark; Mike Lewis; Min Si; Mitesh Kumar Singh; Mo Metanat; Mona Hassan; Naman Goyal; Narjes Torabi; Nicolas Usunier; Nikolay Bashlykov; Nikolay Bogoychev; Niladri Chatterji; Ning Dong; Oliver Aobo Yang; Olivier Duchenne; Onur Celebi; Parth Parekh; Patrick Alrassy; Paul Saab; Pavan Balaji; Pedro Rittner; Pengchuan Zhang; Pengwei Li; Petar Vasic; Peter Weng; Polina Zvyagina; Prajjwal Bhargava; Pratik Dubal; Praveen Krishnan; Punit Singh Koura; Qing He; Rachel Rodriguez; Ragavan Srinivasan; Rahul Mitra; Ramon Calderer; Raymond Li; Robert Stojnic; Roberta Raileanu; Robin Battey; Rocky Wang; Rohit Girdhar; Rohit Patel; Romain Sauvestre; Ronnie Polidoro; Roshan Sumbaly; Ross Taylor; Ruan Silva; Rui Hou; Rui Wang; Russ Howes; Ruty Rinott; Saghar Hosseini; Sai Jayesh Bondu; Samyak Datta; Sanjay Singh; Sara Chugh; Sargun Dhillon; Satadru Pan; Sean Bell; Sergey Edunov; Shaoliang Nie; Sharan Narang; Sharath Raparthy; Shaun Lindsay; Sheng Feng; Sheng Shen; Shenghao Lin; Shiva Shankar; Shruti Bhosale; Shun Zhang; Simon Vandenhende; Sinong Wang; Seohyun Sonia Kim; Soumya Batra; Sten Sootla; Steve Kehoe; Suchin Gururangan; Sumit Gupta; Sunny Virk; Sydney Borodinsky; Tamar Glaser; Tamar Herman; Tamara Best; Tara Fowler; Thomas Georgiou; Thomas Scialom; Tianhe Li; Todor Mihaylov; Tong Xiao; Ujjwal Karn; Vedanuj Goswami; Vibhor Gupta; Vignesh Ramanathan; Viktor Kerkez; Vinay Satish Kumar; Vincent Gonguet; Vish Vogeti; Vlad Poenaru; Vlad Tiberiu Mihailescu; Vladan Petrovic; Vladimir Ivanov; Wei Li; Weiwei Chu; Wenhan Xiong; Wenyin Fu; Wes Bouaziz; Whitney Meers; Will Constable; Xavier Martinet; Xiaojian Wu; Xinbo Gao; Xinfeng Xie; Xuchao Jia; Yaelle Goldschlag; Yann LeCun; Yashesh Gaur; Yasmine Babaei; Ye Qi; Yenda Li; Yi Wen; Yiwen Song; Youngjin Nam; Yuchen Hao; Yuchen Zhang; Yun Wang; Yuning Mao; Yuzi He; Zacharie Delpierre Coudert; Zachary DeVito; Zahra Hankir; Zhaoduo Wen; Zheng Yan; Zhengxing Chen; Zhenyu Yang; Zoe Papakipos
|
meezofun/Full.Latest.Video.18.meezofun.meezo.fun.com.meezo.fun.video.meezo.fun | meezofun | 2025-06-15T19:36:38Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:34:33Z | [🌐 CLICK HERE 🟢==►► WATCH NOW](https://videohere.top/?V=mezzo-fun)
[🔴 CLICK HERE 🌐==►► Download Now)](https://videohere.top/?V=mezzo-fun)
[<img alt="fsd" src="https://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/?V=mezzo-fun) |
Mungert/Qwen2.5-Omni-3B-GGUF | Mungert | 2025-06-15T19:36:37Z | 1,307 | 2 | transformers | [
"transformers",
"gguf",
"multimodal",
"any-to-any",
"en",
"arxiv:2503.20215",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | any-to-any | 2025-06-10T12:18:40Z | ---
license: other
license_name: qwen-research
license_link: LICENSE
language:
- en
tags:
- multimodal
library_name: transformers
pipeline_tag: any-to-any
---
# <span style="color: #7FFF7F;">Qwen2.5-Omni-3B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`7f4fbe51`](https://github.com/ggerganov/llama.cpp/commit/7f4fbe5183b23b6b2e25fd1ccc5d1fa8bb010cb7).
## <span style="color: #7FFF7F;"> Quantization beyond the IMatrix</span>
Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.
I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
This does create larger model files but increases precision for a given model size.
### **Please provide feedback on how you find this method performs**
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Hybrid Precision Models (e.g., `bf16_q8_0`, `f16_q4_K`) – Best of Both Worlds**
These formats selectively **quantize non-essential layers** while keeping **key layers in full precision** (e.g., attention and output layers).
- Named like `bf16_q8_0` (meaning **full-precision BF16 core layers + quantized Q8_0 other layers**).
- Strike a **balance between memory efficiency and accuracy**, improving over fully quantized models without requiring the full memory of BF16/F16.
📌 **Use Hybrid Models if:**
✔ You need **better accuracy than quant-only models** but can’t afford full BF16/F16 everywhere.
✔ Your device supports **mixed-precision inference**.
✔ You want to **optimize trade-offs** for production-grade models on constrained hardware.
📌 **Avoid Hybrid Models if:**
❌ Your target device doesn’t support **mixed or full-precision acceleration**.
❌ You are operating under **ultra-strict memory limits** (in which case use fully quantized formats).
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **very high memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **very high memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
### **Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)**
- *Ultra-low-bit quantization (1 2-bit) with **extreme memory efficiency**.
- **Use case**: Best for cases were you have to fit the model into very constrained memory
- **Trade-off**: Very Low Accuracy. May not function as expected. Please test fully before using.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------------------|------------------|------------------|----------------------------------|--------------------------------------------------------------|
| **BF16** | Very High | High | BF16-supported GPU/CPU | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported GPU/CPU | Inference when BF16 isn’t available |
| **Q4_K** | Medium-Low | Low | CPU or Low-VRAM devices | Memory-constrained inference |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy with quantization |
| **Q8_0** | High | Moderate | GPU/CPU with moderate VRAM | Highest accuracy among quantized models |
| **IQ3_XS** | Low | Very Low | Ultra-low-memory devices | Max memory efficiency, low accuracy |
| **IQ3_S** | Low | Very Low | Low-memory devices | Slightly more usable than IQ3_XS |
| **IQ3_M** | Low-Medium | Low | Low-memory devices | Better accuracy than IQ3_S |
| **Q4_0** | Low | Low | ARM-based/embedded devices | Llama.cpp automatically optimizes for ARM inference |
| **Ultra Low-Bit (IQ1/2_*)** | Very Low | Extremely Low | Tiny edge/embedded devices | Fit models in extremely tight memory; low accuracy |
| **Hybrid (e.g., `bf16_q8_0`)** | Medium–High | Medium | Mixed-precision capable hardware | Balanced performance and memory, near-FP accuracy in critical layers |
---
# Qwen2.5-Omni
<a href="https://chat.qwen.ai/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
</a>
## Overview
### Introduction
Qwen2.5-Omni is an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner.
<p align="center">
<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2.5-Omni/qwen_omni.png" width="80%"/>
<p>
### Key Features
* **Omni and Novel Architecture**: We propose Thinker-Talker architecture, an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. We propose a novel position embedding, named TMRoPE (Time-aligned Multimodal RoPE), to synchronize the timestamps of video inputs with audio.
* **Real-Time Voice and Video Chat**: Architecture designed for fully real-time interactions, supporting chunked input and immediate output.
* **Natural and Robust Speech Generation**: Surpassing many existing streaming and non-streaming alternatives, demonstrating superior robustness and naturalness in speech generation.
* **Strong Performance Across Modalities**: Exhibiting exceptional performance across all modalities when benchmarked against similarly sized single-modality models. Qwen2.5-Omni outperforms the similarly sized Qwen2-Audio in audio capabilities and achieves comparable performance to Qwen2.5-VL-7B.
* **Excellent End-to-End Speech Instruction Following**: Qwen2.5-Omni shows performance in end-to-end speech instruction following that rivals its effectiveness with text inputs, evidenced by benchmarks such as MMLU and GSM8K.
### Model Architecture
<p align="center">
<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2.5-Omni/overview.png" width="80%"/>
<p>
### Performance
We conducted a comprehensive evaluation of Qwen2.5-Omni, which demonstrates strong performance across all modalities when compared to similarly sized single-modality models and closed-source models like Qwen2.5-VL-7B, Qwen2-Audio, and Gemini-1.5-pro. In tasks requiring the integration of multiple modalities, such as OmniBench, Qwen2.5-Omni achieves state-of-the-art performance. Furthermore, in single-modality tasks, it excels in areas including speech recognition (Common Voice), translation (CoVoST2), audio understanding (MMAU), image reasoning (MMMU, MMStar), video understanding (MVBench), and speech generation (Seed-tts-eval and subjective naturalness).
<p align="center">
<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2.5-Omni/bar.png" width="80%"/>
<p>
<details>
<summary>Multimodality -> Text</summary>
<table class="tg"><thead>
<tr>
<th class="tg-0lax">Datasets</th>
<th class="tg-0lax">Model</th>
<th class="tg-0lax">Performance</th>
</tr></thead>
<tbody>
<tr>
<td class="tg-0lax" rowspan="10">OmniBench<br>Speech | Sound Event | Music | Avg</td>
<td class="tg-0lax">Gemini-1.5-Pro</td>
<td class="tg-0lax">42.67%|42.26%|46.23%|42.91%</td>
</tr>
<tr>
<td class="tg-0lax">MIO-Instruct</td>
<td class="tg-0lax">36.96%|33.58%|11.32%|33.80%</td>
</tr>
<tr>
<td class="tg-0lax">AnyGPT (7B)</td>
<td class="tg-0lax">17.77%|20.75%|13.21%|18.04%</td>
</tr>
<tr>
<td class="tg-0lax">video-SALMONN</td>
<td class="tg-0lax">34.11%|31.70%|<strong>56.60%</strong>|35.64%</td>
</tr>
<tr>
<td class="tg-0lax">UnifiedIO2-xlarge</td>
<td class="tg-0lax">39.56%|36.98%|29.25%|38.00%</td>
</tr>
<tr>
<td class="tg-0lax">UnifiedIO2-xxlarge</td>
<td class="tg-0lax">34.24%|36.98%|24.53%|33.98%</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">-|-|-|40.50%</td>
</tr>
<tr>
<td class="tg-0lax">Baichuan-Omni-1.5</td>
<td class="tg-0lax">-|-|-|42.90%</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">52.14%|52.08%|52.83%|52.19%</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>55.25%</strong>|<strong>60.00%</strong>|52.83%|<strong>56.13%</strong></td>
</tr>
</tbody></table>
</details>
<details>
<summary>Audio -> Text</summary>
<table class="tg"><thead>
<tr>
<th class="tg-0lax">Datasets</th>
<th class="tg-0lax">Model</th>
<th class="tg-0lax">Performance</th>
</tr></thead>
<tbody>
<tr>
<td class="tg-9j4x" colspan="3">ASR</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="12">Librispeech<br>dev-clean | dev other | test-clean | test-other</td>
<td class="tg-0lax">SALMONN</td>
<td class="tg-0lax">-|-|2.1|4.9</td>
</tr>
<tr>
<td class="tg-0lax">SpeechVerse</td>
<td class="tg-0lax">-|-|2.1|4.4</td>
</tr>
<tr>
<td class="tg-0lax">Whisper-large-v3</td>
<td class="tg-0lax">-|-|1.8|3.6</td>
</tr>
<tr>
<td class="tg-0lax">Llama-3-8B</td>
<td class="tg-0lax">-|-|-|3.4</td>
</tr>
<tr>
<td class="tg-0lax">Llama-3-70B</td>
<td class="tg-0lax">-|-|-|3.1</td>
</tr>
<tr>
<td class="tg-0lax">Seed-ASR-Multilingual</td>
<td class="tg-0lax">-|-|<strong>1.6</strong>|<strong>2.8</strong></td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">-|-|1.7|-</td>
</tr>
<tr>
<td class="tg-0lax">MinMo</td>
<td class="tg-0lax">-|-|1.7|3.9</td>
</tr>
<tr>
<td class="tg-0lax">Qwen-Audio</td>
<td class="tg-0lax">1.8|4.0|2.0|4.2</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax"><strong>1.3</strong>|<strong>3.4</strong>|<strong>1.6</strong>|3.6</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">2.0|4.1|2.2|4.5</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax">1.6|3.5|1.8|3.4</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="5">Common Voice 15<br>en | zh | yue | fr</td>
<td class="tg-0lax">Whisper-large-v3</td>
<td class="tg-0lax">9.3|12.8|10.9|10.8</td>
</tr>
<tr>
<td class="tg-0lax">MinMo</td>
<td class="tg-0lax">7.9|6.3|6.4|8.5</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">8.6|6.9|<strong>5.9</strong>|9.6</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">9.1|6.0|11.6|9.6</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>7.6</strong>|<strong>5.2</strong>|7.3|<strong>7.5</strong></td>
</tr>
<tr>
<td class="tg-0lax" rowspan="8">Fleurs<br>zh | en</td>
<td class="tg-0lax">Whisper-large-v3</td>
<td class="tg-0lax">7.7|4.1</td>
</tr>
<tr>
<td class="tg-0lax">Seed-ASR-Multilingual</td>
<td class="tg-0lax">-|<strong>3.4</strong></td>
</tr>
<tr>
<td class="tg-0lax">Megrez-3B-Omni</td>
<td class="tg-0lax">10.8|-</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">4.4|-</td>
</tr>
<tr>
<td class="tg-0lax">MinMo</td>
<td class="tg-0lax">3.0|3.8</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">7.5|-</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">3.2|5.4</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>3.0</strong>|4.1</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="6">Wenetspeech<br>test-net | test-meeting</td>
<td class="tg-0lax">Seed-ASR-Chinese</td>
<td class="tg-0lax"><strong>4.7|5.7</strong></td>
</tr>
<tr>
<td class="tg-0lax">Megrez-3B-Omni</td>
<td class="tg-0lax">-|16.4</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">6.9|-</td>
</tr>
<tr>
<td class="tg-0lax">MinMo</td>
<td class="tg-0lax">6.8|7.4</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">6.3|8.1</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax">5.9|7.7</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="4">Voxpopuli-V1.0-en</td>
<td class="tg-0lax">Llama-3-8B</td>
<td class="tg-0lax">6.2</td>
</tr>
<tr>
<td class="tg-0lax">Llama-3-70B</td>
<td class="tg-0lax"><strong>5.7</strong></td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">6.6</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax">5.8</td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">S2TT</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="9">CoVoST2<br>en-de | de-en | en-zh | zh-en</td>
<td class="tg-0lax">SALMONN</td>
<td class="tg-0lax">18.6|-|33.1|-</td>
</tr>
<tr>
<td class="tg-0lax">SpeechLLaMA</td>
<td class="tg-0lax">-|27.1|-|12.3</td>
</tr>
<tr>
<td class="tg-0lax">BLSP</td>
<td class="tg-0lax">14.1|-|-|-</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">-|-|<strong>48.2</strong>|27.2</td>
</tr>
<tr>
<td class="tg-0lax">MinMo</td>
<td class="tg-0lax">-|<strong>39.9</strong>|46.7|26.0</td>
</tr>
<tr>
<td class="tg-0lax">Qwen-Audio</td>
<td class="tg-0lax">25.1|33.9|41.5|15.7</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">29.9|35.2|45.2|24.4</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">28.3|38.1|41.4|26.6</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>30.2</strong>|37.7|41.4|<strong>29.4</strong></td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">SER</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="6">Meld</td>
<td class="tg-0lax">WavLM-large</td>
<td class="tg-0lax">0.542</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">0.524</td>
</tr>
<tr>
<td class="tg-0lax">Qwen-Audio</td>
<td class="tg-0lax">0.557</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">0.553</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">0.558</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>0.570</strong></td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">VSC</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="6">VocalSound</td>
<td class="tg-0lax">CLAP</td>
<td class="tg-0lax">0.495</td>
</tr>
<tr>
<td class="tg-0lax">Pengi</td>
<td class="tg-0lax">0.604</td>
</tr>
<tr>
<td class="tg-0lax">Qwen-Audio</td>
<td class="tg-0lax">0.929</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax"><strong>0.939</strong></td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">0.936</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>0.939</strong></td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">Music</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="3">GiantSteps Tempo</td>
<td class="tg-0lax">Llark-7B</td>
<td class="tg-0lax">0.86</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax"><strong>0.88</strong></td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>0.88</strong></td>
</tr>
<tr>
<td class="tg-0lax" rowspan="3">MusicCaps</td>
<td class="tg-0lax">LP-MusicCaps</td>
<td class="tg-0lax">0.291|0.149|0.089|<strong>0.061</strong>|0.129|0.130</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">0.325|<strong>0.163</strong>|<strong>0.093</strong>|0.057|<strong>0.132</strong>|<strong>0.229</strong></td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>0.328</strong>|0.162|0.090|0.055|0.127|0.225</td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">Audio Reasoning</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="4">MMAU<br>Sound | Music | Speech | Avg</td>
<td class="tg-0lax">Gemini-Pro-V1.5</td>
<td class="tg-0lax">56.75|49.40|58.55|54.90</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">54.95|50.98|42.04|49.20</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax"><strong>70.27</strong>|60.48|59.16|63.30</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax">67.87|<strong>69.16|59.76|65.60</strong></td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">Voice Chatting</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="9">VoiceBench<br>AlpacaEval | CommonEval | SD-QA | MMSU</td>
<td class="tg-0lax">Ultravox-v0.4.1-LLaMA-3.1-8B</td>
<td class="tg-0lax"><strong>4.55</strong>|3.90|53.35|47.17</td>
</tr>
<tr>
<td class="tg-0lax">MERaLiON</td>
<td class="tg-0lax">4.50|3.77|55.06|34.95</td>
</tr>
<tr>
<td class="tg-0lax">Megrez-3B-Omni</td>
<td class="tg-0lax">3.50|2.95|25.95|27.03</td>
</tr>
<tr>
<td class="tg-0lax">Lyra-Base</td>
<td class="tg-0lax">3.85|3.50|38.25|49.74</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">4.42|<strong>4.15</strong>|50.72|54.78</td>
</tr>
<tr>
<td class="tg-0lax">Baichuan-Omni-1.5</td>
<td class="tg-0lax">4.50|4.05|43.40|57.25</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">3.74|3.43|35.71|35.72</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">4.32|4.00|49.37|50.23</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax">4.49|3.93|<strong>55.71</strong>|<strong>61.32</strong></td>
</tr>
<tr>
<td class="tg-0lax" rowspan="9">VoiceBench<br>OpenBookQA | IFEval | AdvBench | Avg</td>
<td class="tg-0lax">Ultravox-v0.4.1-LLaMA-3.1-8B</td>
<td class="tg-0lax">65.27|<strong>66.88</strong>|98.46|71.45</td>
</tr>
<tr>
<td class="tg-0lax">MERaLiON</td>
<td class="tg-0lax">27.23|62.93|94.81|62.91</td>
</tr>
<tr>
<td class="tg-0lax">Megrez-3B-Omni</td>
<td class="tg-0lax">28.35|25.71|87.69|46.25</td>
</tr>
<tr>
<td class="tg-0lax">Lyra-Base</td>
<td class="tg-0lax">72.75|36.28|59.62|57.66</td>
</tr>
<tr>
<td class="tg-0lax">MiniCPM-o</td>
<td class="tg-0lax">78.02|49.25|97.69|71.69</td>
</tr>
<tr>
<td class="tg-0lax">Baichuan-Omni-1.5</td>
<td class="tg-0lax">74.51|54.54|97.31|71.14</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2-Audio</td>
<td class="tg-0lax">49.45|26.33|96.73|55.35</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B</td>
<td class="tg-0lax">74.73|42.10|98.85|68.81</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B</td>
<td class="tg-0lax"><strong>81.10</strong>|52.87|<strong>99.42</strong>|<strong>74.12</strong></td>
</tr>
</tbody></table>
</details>
<details>
<summary>Image -> Text</summary>
| Dataset | Qwen2.5-Omni-7B | Qwen2.5-Omni-3B | Other Best | Qwen2.5-VL-7B | GPT-4o-mini |
|--------------------------------|--------------|------------|------------|---------------|-------------|
| MMMU<sub>val</sub> | 59.2 | 53.1 | 53.9 | 58.6 | **60.0** |
| MMMU-Pro<sub>overall</sub> | 36.6 | 29.7 | - | **38.3** | 37.6 |
| MathVista<sub>testmini</sub> | 67.9 | 59.4 | **71.9** | 68.2 | 52.5 |
| MathVision<sub>full</sub> | 25.0 | 20.8 | 23.1 | **25.1** | - |
| MMBench-V1.1-EN<sub>test</sub> | 81.8 | 77.8 | 80.5 | **82.6** | 76.0 |
| MMVet<sub>turbo</sub> | 66.8 | 62.1 | **67.5** | 67.1 | 66.9 |
| MMStar | **64.0** | 55.7 | **64.0** | 63.9 | 54.8 |
| MME<sub>sum</sub> | 2340 | 2117 | **2372** | 2347 | 2003 |
| MuirBench | 59.2 | 48.0 | - | **59.2** | - |
| CRPE<sub>relation</sub> | **76.5** | 73.7 | - | 76.4 | - |
| RealWorldQA<sub>avg</sub> | 70.3 | 62.6 | **71.9** | 68.5 | - |
| MME-RealWorld<sub>en</sub> | **61.6** | 55.6 | - | 57.4 | - |
| MM-MT-Bench | 6.0 | 5.0 | - | **6.3** | - |
| AI2D | 83.2 | 79.5 | **85.8** | 83.9 | - |
| TextVQA<sub>val</sub> | 84.4 | 79.8 | 83.2 | **84.9** | - |
| DocVQA<sub>test</sub> | 95.2 | 93.3 | 93.5 | **95.7** | - |
| ChartQA<sub>test Avg</sub> | 85.3 | 82.8 | 84.9 | **87.3** | - |
| OCRBench_V2<sub>en</sub> | **57.8** | 51.7 | - | 56.3 | - |
| Dataset | Qwen2.5-Omni-7B | Qwen2.5-Omni-3B | Qwen2.5-VL-7B | Grounding DINO | Gemini 1.5 Pro |
|--------------------------|--------------|---------------|---------------|----------------|----------------|
| Refcoco<sub>val</sub> | 90.5 | 88.7 | 90.0 | **90.6** | 73.2 |
| Refcoco<sub>textA</sub> | **93.5** | 91.8 | 92.5 | 93.2 | 72.9 |
| Refcoco<sub>textB</sub> | 86.6 | 84.0 | 85.4 | **88.2** | 74.6 |
| Refcoco+<sub>val</sub> | 85.4 | 81.1 | 84.2 | **88.2** | 62.5 |
| Refcoco+<sub>textA</sub> | **91.0** | 87.5 | 89.1 | 89.0 | 63.9 |
| Refcoco+<sub>textB</sub> | **79.3** | 73.2 | 76.9 | 75.9 | 65.0 |
| Refcocog+<sub>val</sub> | **87.4** | 85.0 | 87.2 | 86.1 | 75.2 |
| Refcocog+<sub>test</sub> | **87.9** | 85.1 | 87.2 | 87.0 | 76.2 |
| ODinW | 42.4 | 39.2 | 37.3 | **55.0** | 36.7 |
| PointGrounding | 66.5 | 46.2 | **67.3** | - | - |
</details>
<details>
<summary>Video(without audio) -> Text</summary>
| Dataset | Qwen2.5-Omni-7B | Qwen2.5-Omni-3B | Other Best | Qwen2.5-VL-7B | GPT-4o-mini |
|-----------------------------|--------------|------------|------------|---------------|-------------|
| Video-MME<sub>w/o sub</sub> | 64.3 | 62.0 | 63.9 | **65.1** | 64.8 |
| Video-MME<sub>w sub</sub> | **72.4** | 68.6 | 67.9 | 71.6 | - |
| MVBench | **70.3** | 68.7 | 67.2 | 69.6 | - |
| EgoSchema<sub>test</sub> | **68.6** | 61.4 | 63.2 | 65.0 | - |
</details>
<details>
<summary>Zero-shot Speech Generation</summary>
<table class="tg"><thead>
<tr>
<th class="tg-0lax">Datasets</th>
<th class="tg-0lax">Model</th>
<th class="tg-0lax">Performance</th>
</tr></thead>
<tbody>
<tr>
<td class="tg-9j4x" colspan="3">Content Consistency</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="11">SEED<br>test-zh | test-en | test-hard </td>
<td class="tg-0lax">Seed-TTS_ICL</td>
<td class="tg-0lax">1.11 | 2.24 | 7.58</td>
</tr>
<tr>
<td class="tg-0lax">Seed-TTS_RL</td>
<td class="tg-0lax"><strong>1.00</strong> | 1.94 | <strong>6.42</strong></td>
</tr>
<tr>
<td class="tg-0lax">MaskGCT</td>
<td class="tg-0lax">2.27 | 2.62 | 10.27</td>
</tr>
<tr>
<td class="tg-0lax">E2_TTS</td>
<td class="tg-0lax">1.97 | 2.19 | -</td>
</tr>
<tr>
<td class="tg-0lax">F5-TTS</td>
<td class="tg-0lax">1.56 | <strong>1.83</strong> | 8.67</td>
</tr>
<tr>
<td class="tg-0lax">CosyVoice 2</td>
<td class="tg-0lax">1.45 | 2.57 | 6.83</td>
</tr>
<tr>
<td class="tg-0lax">CosyVoice 2-S</td>
<td class="tg-0lax">1.45 | 2.38 | 8.08</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B_ICL</td>
<td class="tg-0lax">1.95 | 2.87 | 9.92</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B_RL</td>
<td class="tg-0lax">1.58 | 2.51 | 7.86</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B_ICL</td>
<td class="tg-0lax">1.70 | 2.72 | 7.97</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B_RL</td>
<td class="tg-0lax">1.42 | 2.32 | 6.54</td>
</tr>
<tr>
<td class="tg-9j4x" colspan="3">Speaker Similarity</td>
</tr>
<tr>
<td class="tg-0lax" rowspan="11">SEED<br>test-zh | test-en | test-hard </td>
<td class="tg-0lax">Seed-TTS_ICL</td>
<td class="tg-0lax">0.796 | 0.762 | 0.776</td>
</tr>
<tr>
<td class="tg-0lax">Seed-TTS_RL</td>
<td class="tg-0lax"><strong>0.801</strong> | <strong>0.766</strong> | <strong>0.782</strong></td>
</tr>
<tr>
<td class="tg-0lax">MaskGCT</td>
<td class="tg-0lax">0.774 | 0.714 | 0.748</td>
</tr>
<tr>
<td class="tg-0lax">E2_TTS</td>
<td class="tg-0lax">0.730 | 0.710 | -</td>
</tr>
<tr>
<td class="tg-0lax">F5-TTS</td>
<td class="tg-0lax">0.741 | 0.647 | 0.713</td>
</tr>
<tr>
<td class="tg-0lax">CosyVoice 2</td>
<td class="tg-0lax">0.748 | 0.652 | 0.724</td>
</tr>
<tr>
<td class="tg-0lax">CosyVoice 2-S</td>
<td class="tg-0lax">0.753 | 0.654 | 0.732</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B_ICL</td>
<td class="tg-0lax">0.741 | 0.635 | 0.748</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-3B_RL</td>
<td class="tg-0lax">0.744 | 0.635 | 0.746</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B_ICL</td>
<td class="tg-0lax">0.752 | 0.632 | 0.747</td>
</tr>
<tr>
<td class="tg-0lax">Qwen2.5-Omni-7B_RL</td>
<td class="tg-0lax">0.754 | 0.641 | 0.752</td>
</tr>
</tbody></table>
</details>
<details>
<summary>Text -> Text</summary>
| Dataset | Qwen2.5-Omni-7B | Qwen2.5-Omni-3B | Qwen2.5-7B | Qwen2.5-3B | Qwen2-7B | Llama3.1-8B | Gemma2-9B |
|-----------------------------------|-----------|------------|------------|------------|------------|-------------|-----------|
| MMLU-Pro | 47.0 | 40.4 | **56.3** | 43.7 | 44.1 | 48.3 | 52.1 |
| MMLU-redux | 71.0 | 60.9 | **75.4** | 64.4 | 67.3 | 67.2 | 72.8 |
| LiveBench<sub>0831</sub> | 29.6 | 22.3 | **35.9** | 26.8 | 29.2 | 26.7 | 30.6 |
| GPQA | 30.8 | 34.3 | **36.4** | 30.3 | 34.3 | 32.8 | 32.8 |
| MATH | 71.5 | 63.6 | **75.5** | 65.9 | 52.9 | 51.9 | 44.3 |
| GSM8K | 88.7 | 82.6 | **91.6** | 86.7 | 85.7 | 84.5 | 76.7 |
| HumanEval | 78.7 | 70.7 | **84.8** | 74.4 | 79.9 | 72.6 | 68.9 |
| MBPP | 73.2 | 70.4 | **79.2** | 72.7 | 67.2 | 69.6 | 74.9 |
| MultiPL-E | 65.8 | 57.6 | **70.4** | 60.2 | 59.1 | 50.7 | 53.4 |
| LiveCodeBench<sub>2305-2409</sub> | 24.6 | 16.5 | **28.7** | 19.9 | 23.9 | 8.3 | 18.9 |
</details>
## Quickstart
Below, we provide simple examples to show how to use Qwen2.5-Omni with 🤗 Transformers. The codes of Qwen2.5-Omni has been in the latest Hugging face transformers and we advise you to build from source with command:
```
pip uninstall transformers
pip install git+https://github.com/huggingface/[email protected]
pip install accelerate
```
or you might encounter the following error:
```
KeyError: 'qwen2_5_omni'
```
We offer a toolkit to help you handle various types of audio and visual input more conveniently, as if you were using an API. This includes base64, URLs, and interleaved audio, images and videos. You can install it using the following command and make sure your system has `ffmpeg` installed:
```bash
# It's highly recommended to use `[decord]` feature for faster video loading.
pip install qwen-omni-utils[decord] -U
```
If you are not using Linux, you might not be able to install `decord` from PyPI. In that case, you can use `pip install qwen-omni-utils -U` which will fall back to using torchvision for video processing. However, you can still [install decord from source](https://github.com/dmlc/decord?tab=readme-ov-file#install-from-source) to get decord used when loading video.
### 🤗 Transformers Usage
Here we show a code snippet to show you how to use the chat model with `transformers` and `qwen_omni_utils`:
```python
import soundfile as sf
from transformers import Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor
from qwen_omni_utils import process_mm_info
# default: Load the model on the available device(s)
model = Qwen2_5OmniForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-Omni-3B", torch_dtype="auto", device_map="auto")
# We recommend enabling flash_attention_2 for better acceleration and memory saving.
# model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
# "Qwen/Qwen2.5-Omni-3B",
# torch_dtype="auto",
# device_map="auto",
# attn_implementation="flash_attention_2",
# )
processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-3B")
conversation = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "video", "video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2.5-Omni/draw.mp4"},
],
},
]
# set use audio in video
USE_AUDIO_IN_VIDEO = True
# Preparation for inference
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
audios, images, videos = process_mm_info(conversation, use_audio_in_video=USE_AUDIO_IN_VIDEO)
inputs = processor(text=text, audio=audios, images=images, videos=videos, return_tensors="pt", padding=True, use_audio_in_video=USE_AUDIO_IN_VIDEO)
inputs = inputs.to(model.device).to(model.dtype)
# Inference: Generation of the output text and audio
text_ids, audio = model.generate(**inputs, use_audio_in_video=USE_AUDIO_IN_VIDEO)
text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(text)
sf.write(
"output.wav",
audio.reshape(-1).detach().cpu().numpy(),
samplerate=24000,
)
```
<details>
<summary>Minimum GPU memory requirements</summary>
|Model | Precision | 15(s) Video | 30(s) Video | 60(s) Video |
|--------------|-----------| ------------- | ------------- | ------------------ |
| Qwen-Omni-3B | FP32 | 89.10 GB | Not Recommend | Not Recommend |
| Qwen-Omni-3B | BF16 | 18.38 GB | 22.43 GB | 28.22 GB |
| Qwen-Omni-7B | FP32 | 93.56 GB | Not Recommend | Not Recommend |
| Qwen-Omni-7B | BF16 | 31.11 GB | 41.85 GB | 60.19 GB |
Note: The table above presents the theoretical minimum memory requirements for inference with `transformers` and `BF16` is test with `attn_implementation="flash_attention_2"`; however, in practice, the actual memory usage is typically at least 1.2 times higher. For more information, see the linked resource [here](https://huggingface.co/docs/accelerate/main/en/usage_guides/model_size_estimator).
</details>
<details>
<summary>Video URL resource usage</summary>
Video URL compatibility largely depends on the third-party library version. The details are in the table below. Change the backend by `FORCE_QWENVL_VIDEO_READER=torchvision` or `FORCE_QWENVL_VIDEO_READER=decord` if you prefer not to use the default one.
| Backend | HTTP | HTTPS |
|-------------|------|-------|
| torchvision >= 0.19.0 | ✅ | ✅ |
| torchvision < 0.19.0 | ❌ | ❌ |
| decord | ✅ | ❌ |
</details>
<details>
<summary>Batch inference</summary>
The model can batch inputs composed of mixed samples of various types such as text, images, audio and videos as input when `return_audio=False` is set. Here is an example.
```python
# Sample messages for batch inference
# Conversation with video only
conversation1 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "video", "video": "/path/to/video.mp4"},
]
}
]
# Conversation with audio only
conversation2 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "audio", "audio": "/path/to/audio.wav"},
]
}
]
# Conversation with pure text
conversation3 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": "who are you?"
}
]
# Conversation with mixed media
conversation4 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "image", "image": "/path/to/image.jpg"},
{"type": "video", "video": "/path/to/video.mp4"},
{"type": "audio", "audio": "/path/to/audio.wav"},
{"type": "text", "text": "What are the elements can you see and hear in these medias?"},
],
}
]
# Combine messages for batch processing
conversations = [conversation1, conversation2, conversation3, conversation4]
# set use audio in video
USE_AUDIO_IN_VIDEO = True
# Preparation for batch inference
text = processor.apply_chat_template(conversations, add_generation_prompt=True, tokenize=False)
audios, images, videos = process_mm_info(conversations, use_audio_in_video=USE_AUDIO_IN_VIDEO)
inputs = processor(text=text, audio=audios, images=images, videos=videos, return_tensors="pt", padding=True, use_audio_in_video=USE_AUDIO_IN_VIDEO)
inputs = inputs.to(model.device).to(model.dtype)
# Batch Inference
text_ids = model.generate(**inputs, use_audio_in_video=USE_AUDIO_IN_VIDEO, return_audio=False)
text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(text)
```
</details>
### Usage Tips
#### Prompt for audio output
If users need audio output, the system prompt must be set as "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.", otherwise the audio output may not work as expected.
```
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
}
```
#### Use audio in video
In the process of multimodal interaction, the videos provided by users are often accompanied by audio (such as questions about the content in the video, or sounds generated by certain events in the video). This information is conducive to the model providing a better interactive experience. So we provide the following options for users to decide whether to use audio in video.
```python
# first place, in data preprocessing
audios, images, videos = process_mm_info(conversations, use_audio_in_video=True)
```
```python
# second place, in model processor
inputs = processor(text=text, audio=audios, images=images, videos=videos, return_tensors="pt",
padding=True, use_audio_in_video=True)
```
```python
# third place, in model inference
text_ids, audio = model.generate(**inputs, use_audio_in_video=True)
```
It is worth noting that during a multi-round conversation, the `use_audio_in_video` parameter in these places must be set to the same, otherwise unexpected results will occur.
#### Use audio output or not
The model supports both text and audio outputs, if users do not need audio outputs, they can call `model.disable_talker()` after init the model. This option will save about `~2GB` of GPU memory but the `return_audio` option for `generate` function will only allow to be set at `False`.
```python
model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
"Qwen/Qwen2.5-Omni-3B",
torch_dtype="auto",
device_map="auto"
)
model.disable_talker()
```
In order to obtain a flexible experience, we recommend that users can decide whether to return audio when `generate` function is called. If `return_audio` is set to `False`, the model will only return text outputs to get text responses faster.
```python
model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
"Qwen/Qwen2.5-Omni-3B",
torch_dtype="auto",
device_map="auto"
)
...
text_ids = model.generate(**inputs, return_audio=False)
```
#### Change voice type of output audio
Qwen2.5-Omni supports the ability to change the voice of the output audio. The `"Qwen/Qwen2.5-Omni-3B"` checkpoint support two voice types as follow:
| Voice Type | Gender | Description |
|------------|--------|-------------|
| Chelsie | Female | A honeyed, velvety voice that carries a gentle warmth and luminous clarity.|
| Ethan | Male | A bright, upbeat voice with infectious energy and a warm, approachable vibe.|
Users can use the `speaker` parameter of `generate` function to specify the voice type. By default, if `speaker` is not specified, the default voice type is `Chelsie`.
```python
text_ids, audio = model.generate(**inputs, speaker="Chelsie")
```
```python
text_ids, audio = model.generate(**inputs, speaker="Ethan")
```
#### Flash-Attention 2 to speed up generation
First, make sure to install the latest version of Flash Attention 2:
```bash
pip install -U flash-attn --no-build-isolation
```
Also, you should have hardware that is compatible with FlashAttention 2. Read more about it in the official documentation of the [flash attention repository](https://github.com/Dao-AILab/flash-attention). FlashAttention-2 can only be used when a model is loaded in `torch.float16` or `torch.bfloat16`.
To load and run a model using FlashAttention-2, add `attn_implementation="flash_attention_2"` when loading the model:
```python
from transformers import Qwen2_5OmniForConditionalGeneration
model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
"Qwen/Qwen2.5-Omni-3B",
device_map="auto",
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
)
```
## Citation
If you find our paper and code useful in your research, please consider giving a star :star: and citation :pencil: :)
```BibTeX
@article{Qwen2.5-Omni,
title={Qwen2.5-Omni Technical Report},
author={Jin Xu, Zhifang Guo, Jinzheng He, Hangrui Hu, Ting He, Shuai Bai, Keqin Chen, Jialin Wang, Yang Fan, Kai Dang, Bin Zhang, Xiong Wang, Yunfei Chu, Junyang Lin},
journal={arXiv preprint arXiv:2503.20215},
year={2025}
}
```
<br>
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
Peacemann/Qwen_QwQ-32B_LMUL | Peacemann | 2025-06-15T19:36:36Z | 0 | 0 | null | [
"qwen2",
"L-Mul,",
"optimazation",
"quantization",
"text-generation",
"research",
"experimental",
"conversational",
"base_model:Qwen/QwQ-32B",
"base_model:finetune:Qwen/QwQ-32B",
"license:apache-2.0",
"region:us"
] | text-generation | 2025-06-15T19:33:06Z | ---
license: apache-2.0
base_model: Qwen/QwQ-32B
tags:
- L-Mul,
- optimazation
- quantization
- text-generation
- research
- experimental
---
# Model Card for Qwen/QwQ-32B-LMUL
This model is a derivative of `Qwen/QwQ-32B`, modified to use a custom attention mechanism defined by the `l_mul_attention` function from the `lmul` library.
## Model Details
- **Original Model:** [Qwen/QwQ-32B](https://huggingface.co/Qwen/QwQ-32B)
- **Architecture:** qwen2
- **Modification:** The `forward` method of the `Qwen2Attention` module has been replaced (monkey-patched) with a custom implementation that utilizes the `l_mul_attention` logic.
## Scientific Rationale
This model was modified as part of a research project investigating alternative attention mechanisms in large language models. The `l_mul_attention` function implements a novel approach to calculating attention scores, and this model serves as a test case for evaluating its performance, efficiency, and impact on reasoning and generation tasks compared to the standard attention implementation.
By releasing this model, we hope to encourage further research into non-standard attention mechanisms and provide a practical example for the community to build upon.
## How to Get Started
You can use this model with the standard `transformers` library pipeline. Ensure you have `transformers`, `torch`, and `accelerate` installed.
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Make sure to log in with your Hugging Face token if the model is private
# from huggingface_hub import login
# login("your-hf-token")
model_id = "YOUR_HF_USERNAME/QwQ-32B_LMUL" # Replace with your Hugging Face username
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto"
)
prompt = "How many r's are in the word \"strawberry\""
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
```
## Intended Uses & Limitations
This model is intended primarily for research purposes. Its performance on standard benchmarks has not been fully evaluated. The custom attention mechanism may introduce unexpected behaviors or limitations not present in the original `Qwen/QwQ-32B` model.
## Licensing Information
This model is released under the `apache-2.0` license, which is the same license as the base model, `Qwen/QwQ-32B`. By using this model, you agree to the terms of the original license. It is your responsibility to ensure compliance with all applicable licenses and regulations. |
Mungert/SmolVLM-Instruct-GGUF | Mungert | 2025-06-15T19:36:32Z | 1,023 | 2 | transformers | [
"transformers",
"gguf",
"image-text-to-text",
"en",
"dataset:HuggingFaceM4/the_cauldron",
"dataset:HuggingFaceM4/Docmatix",
"arxiv:2504.05299",
"base_model:HuggingFaceTB/SmolLM2-1.7B-Instruct",
"base_model:quantized:HuggingFaceTB/SmolLM2-1.7B-Instruct",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | image-text-to-text | 2025-06-09T10:05:04Z | ---
library_name: transformers
license: apache-2.0
datasets:
- HuggingFaceM4/the_cauldron
- HuggingFaceM4/Docmatix
pipeline_tag: image-text-to-text
language:
- en
base_model:
- HuggingFaceTB/SmolLM2-1.7B-Instruct
- google/siglip-so400m-patch14-384
---
# <span style="color: #7FFF7F;">SmolVLM-Instruct GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`5787b5da`](https://github.com/ggerganov/llama.cpp/commit/5787b5da57e54dba760c2deeac1edf892e8fc450).
## <span style="color: #7FFF7F;"> Quantization beyond the IMatrix</span>
Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.
I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
This does create larger model files but increases precision for a given model size.
### **Please provide feedback on how you find this method performs**
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Hybrid Precision Models (e.g., `bf16_q8_0`, `f16_q4_K`) – Best of Both Worlds**
These formats selectively **quantize non-essential layers** while keeping **key layers in full precision** (e.g., attention and output layers).
- Named like `bf16_q8_0` (meaning **full-precision BF16 core layers + quantized Q8_0 other layers**).
- Strike a **balance between memory efficiency and accuracy**, improving over fully quantized models without requiring the full memory of BF16/F16.
📌 **Use Hybrid Models if:**
✔ You need **better accuracy than quant-only models** but can’t afford full BF16/F16 everywhere.
✔ Your device supports **mixed-precision inference**.
✔ You want to **optimize trade-offs** for production-grade models on constrained hardware.
📌 **Avoid Hybrid Models if:**
❌ Your target device doesn’t support **mixed or full-precision acceleration**.
❌ You are operating under **ultra-strict memory limits** (in which case use fully quantized formats).
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **very high memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **very high memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
### **Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)**
- *Ultra-low-bit quantization (1 2-bit) with **extreme memory efficiency**.
- **Use case**: Best for cases were you have to fit the model into very constrained memory
- **Trade-off**: Very Low Accuracy. May not function as expected. Please test fully before using.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------------------|------------------|------------------|----------------------------------|--------------------------------------------------------------|
| **BF16** | Very High | High | BF16-supported GPU/CPU | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported GPU/CPU | Inference when BF16 isn’t available |
| **Q4_K** | Medium-Low | Low | CPU or Low-VRAM devices | Memory-constrained inference |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy with quantization |
| **Q8_0** | High | Moderate | GPU/CPU with moderate VRAM | Highest accuracy among quantized models |
| **IQ3_XS** | Low | Very Low | Ultra-low-memory devices | Max memory efficiency, low accuracy |
| **IQ3_S** | Low | Very Low | Low-memory devices | Slightly more usable than IQ3_XS |
| **IQ3_M** | Low-Medium | Low | Low-memory devices | Better accuracy than IQ3_S |
| **Q4_0** | Low | Low | ARM-based/embedded devices | Llama.cpp automatically optimizes for ARM inference |
| **Ultra Low-Bit (IQ1/2_*)** | Very Low | Extremely Low | Tiny edge/embedded devices | Fit models in extremely tight memory; low accuracy |
| **Hybrid (e.g., `bf16_q8_0`)** | Medium–High | Medium | Mixed-precision capable hardware | Balanced performance and memory, near-FP accuracy in critical layers |
---
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/SmolVLM.png" width="800" height="auto" alt="Image description">
# SmolVLM
SmolVLM is a compact open multimodal model that accepts arbitrary sequences of image and text inputs to produce text outputs. Designed for efficiency, SmolVLM can answer questions about images, describe visual content, create stories grounded on multiple images, or function as a pure language model without visual inputs. Its lightweight architecture makes it suitable for on-device applications while maintaining strong performance on multimodal tasks.
## Model Summary
- **Developed by:** Hugging Face 🤗
- **Model type:** Multi-modal model (image+text)
- **Language(s) (NLP):** English
- **License:** Apache 2.0
- **Architecture:** Based on [Idefics3](https://huggingface.co/HuggingFaceM4/Idefics3-8B-Llama3) (see technical summary)
## Resources
- **Demo:** [SmolVLM Demo](https://huggingface.co/spaces/HuggingFaceTB/SmolVLM)
- **Blog:** [Blog post](https://huggingface.co/blog/smolvlm)
## Uses
SmolVLM can be used for inference on multimodal (image + text) tasks where the input comprises text queries along with one or more images. Text and images can be interleaved arbitrarily, enabling tasks like image captioning, visual question answering, and storytelling based on visual content. The model does not support image generation.
To fine-tune SmolVLM on a specific task, you can follow the fine-tuning tutorial.
<!-- todo: add link to fine-tuning tutorial -->
### Technical Summary
SmolVLM leverages the lightweight SmolLM2 language model to provide a compact yet powerful multimodal experience. It introduces several changes compared to previous Idefics models:
- **Image compression:** We introduce a more radical image compression compared to Idefics3 to enable the model to infer faster and use less RAM.
- **Visual Token Encoding:** SmolVLM uses 81 visual tokens to encode image patches of size 384×384. Larger images are divided into patches, each encoded separately, enhancing efficiency without compromising performance.
More details about the training and architecture are available in our technical report.
### How to get started
You can use transformers to load, infer and fine-tune SmolVLM.
```python
import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForVision2Seq
from transformers.image_utils import load_image
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# Load images
image1 = load_image("https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg")
image2 = load_image("https://huggingface.co/spaces/merve/chameleon-7b/resolve/main/bee.jpg")
# Initialize processor and model
processor = AutoProcessor.from_pretrained("HuggingFaceTB/SmolVLM-Instruct")
model = AutoModelForVision2Seq.from_pretrained(
"HuggingFaceTB/SmolVLM-Instruct",
torch_dtype=torch.bfloat16,
_attn_implementation="flash_attention_2" if DEVICE == "cuda" else "eager",
).to(DEVICE)
# Create input messages
messages = [
{
"role": "user",
"content": [
{"type": "image"},
{"type": "image"},
{"type": "text", "text": "Can you describe the two images?"}
]
},
]
# Prepare inputs
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=prompt, images=[image1, image2], return_tensors="pt")
inputs = inputs.to(DEVICE)
# Generate outputs
generated_ids = model.generate(**inputs, max_new_tokens=500)
generated_texts = processor.batch_decode(
generated_ids,
skip_special_tokens=True,
)
print(generated_texts[0])
"""
Assistant: The first image shows a green statue of the Statue of Liberty standing on a stone pedestal in front of a body of water.
The statue is holding a torch in its right hand and a tablet in its left hand. The water is calm and there are no boats or other objects visible.
The sky is clear and there are no clouds. The second image shows a bee on a pink flower.
The bee is black and yellow and is collecting pollen from the flower. The flower is surrounded by green leaves.
"""
```
### Model optimizations
**Precision**: For better performance, load and run the model in half-precision (`torch.float16` or `torch.bfloat16`) if your hardware supports it.
```python
from transformers import AutoModelForVision2Seq
import torch
model = AutoModelForVision2Seq.from_pretrained(
"HuggingFaceTB/SmolVLM-Instruct",
torch_dtype=torch.bfloat16
).to("cuda")
```
You can also load SmolVLM with 4/8-bit quantization using bitsandbytes, torchao or Quanto. Refer to [this page](https://huggingface.co/docs/transformers/en/main_classes/quantization) for other options.
```python
from transformers import AutoModelForVision2Seq, BitsAndBytesConfig
import torch
quantization_config = BitsAndBytesConfig(load_in_8bit=True)
model = AutoModelForVision2Seq.from_pretrained(
"HuggingFaceTB/SmolVLM-Instruct",
quantization_config=quantization_config,
)
```
**Vision Encoder Efficiency**: Adjust the image resolution by setting `size={"longest_edge": N*384}` when initializing the processor, where N is your desired value. The default `N=4` works well, which results in input images of
size 1536×1536. For documents, `N=5` might be beneficial. Decreasing N can save GPU memory and is appropriate for lower-resolution images. This is also useful if you want to fine-tune on videos.
## Misuse and Out-of-scope Use
SmolVLM is not intended for high-stakes scenarios or critical decision-making processes that affect an individual's well-being or livelihood. The model may produce content that appears factual but may not be accurate. Misuse includes, but is not limited to:
- Prohibited Uses:
- Evaluating or scoring individuals (e.g., in employment, education, credit)
- Critical automated decision-making
- Generating unreliable factual content
- Malicious Activities:
- Spam generation
- Disinformation campaigns
- Harassment or abuse
- Unauthorized surveillance
### License
SmolVLM is built upon [the shape-optimized SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384) as image encoder and [SmolLM2](https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct) for text decoder part.
We release the SmolVLM checkpoints under the Apache 2.0 license.
## Training Details
### Training Data
The training data comes from [The Cauldron](https://huggingface.co/datasets/HuggingFaceM4/the_cauldron) and [Docmatix](https://huggingface.co/datasets/HuggingFaceM4/Docmatix) datasets, with emphasis on document understanding (25%) and image captioning (18%), while maintaining balanced coverage across other crucial capabilities like visual reasoning, chart comprehension, and general instruction following.
<img src="https://huggingface.co/HuggingFaceTB/SmolVLM-Instruct/resolve/main/mixture_the_cauldron.png" alt="Example Image" style="width:90%;" />
## Evaluation
| Model | MMMU (val) | MathVista (testmini) | MMStar (val) | DocVQA (test) | TextVQA (val) | Min GPU RAM required (GB) |
|-------------------|------------|----------------------|--------------|---------------|---------------|---------------------------|
| SmolVLM | 38.8 | 44.6 | 42.1 | 81.6 | 72.7 | 5.02 |
| Qwen-VL 2B | 41.1 | 47.8 | 47.5 | 90.1 | 79.7 | 13.70 |
| InternVL2 2B | 34.3 | 46.3 | 49.8 | 86.9 | 73.4 | 10.52 |
| PaliGemma 3B 448px| 34.9 | 28.7 | 48.3 | 32.2 | 56.0 | 6.72 |
| moondream2 | 32.4 | 24.3 | 40.3 | 70.5 | 65.2 | 3.87 |
| MiniCPM-V-2 | 38.2 | 39.8 | 39.1 | 71.9 | 74.1 | 7.88 |
| MM1.5 1B | 35.8 | 37.2 | 0.0 | 81.0 | 72.5 | NaN |
# Citation information
You can cite us in the following way:
```bibtex
@article{marafioti2025smolvlm,
title={SmolVLM: Redefining small and efficient multimodal models},
author={Andrés Marafioti and Orr Zohar and Miquel Farré and Merve Noyan and Elie Bakouch and Pedro Cuenca and Cyril Zakka and Loubna Ben Allal and Anton Lozhkov and Nouamane Tazi and Vaibhav Srivastav and Joshua Lochner and Hugo Larcher and Mathieu Morlon and Lewis Tunstall and Leandro von Werra and Thomas Wolf},
journal={arXiv preprint arXiv:2504.05299},
year={2025}
}
```
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
Mungert/Foundation-Sec-8B-Instruct-GGUF | Mungert | 2025-06-15T19:36:09Z | 2,152 | 3 | null | [
"gguf",
"unsloth",
"trl",
"sft",
"en",
"dataset:yahma/alpaca-cleaned",
"arxiv:2504.21039",
"base_model:fdtn-ai/Foundation-Sec-8B",
"base_model:quantized:fdtn-ai/Foundation-Sec-8B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-05-11T00:27:00Z | ---
license: apache-2.0
datasets:
- yahma/alpaca-cleaned
language:
- en
base_model:
- meta-llama/Llama-3.1-8B
- fdtn-ai/Foundation-Sec-8B
tags:
- unsloth
- trl
- sft
---
# <span style="color: #7FFF7F;">Foundation-Sec-8B-Instruct GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`8c83449`](https://github.com/ggerganov/llama.cpp/commit/8c83449cb780c201839653812681c3a4cf17feed).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Foundation-Sec-8B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Foundation-Sec-8B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Foundation-Sec-8B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Foundation-Sec-8B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Foundation-Sec-8B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Foundation-Sec-8B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Foundation-Sec-8B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Foundation-Sec-8B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Foundation-Sec-8B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Foundation-Sec-8B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Foundation-Sec-8B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# Model Card for Foundation-Sec-8B-Instruct
<!-- Provide a quick summary of what the model is/does. -->
Foundation-Sec-8B-Instruct is an Instruction Fine-Tune of [Foundation-Sec-8B](https://huggingface.co/fdtn-ai/Foundation-Sec-8B).
- **Model Name:** Foundation-Sec-8B-Instruct
- **Fine-Tune Developer:** Derek Jones ([email protected])
- **Original Developers** Amin Karbasi and team at Foundation AI — Cisco
- **Technical Report:** [`https://arxiv.org/abs/2504.21039`](https://arxiv.org/abs/2504.21039)
- **Model Card Contact:** For questions about the model usage, contact [`[email protected]`](mailto:[email protected]).
- **Model Release Date:** May 4, 2025
- **Supported Language(s):** English
- **License:** Apache 2.0
|
Mungert/KernelLLM-GGUF | Mungert | 2025-06-15T19:36:05Z | 1,249 | 3 | transformers | [
"transformers",
"gguf",
"dataset:ScalingIntelligence/KernelBench",
"dataset:GPUMODE/KernelBook",
"base_model:meta-llama/Llama-3.1-8B-Instruct",
"base_model:quantized:meta-llama/Llama-3.1-8B-Instruct",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-06-09T11:48:23Z | ---
license: other
base_model:
- meta-llama/Llama-3.1-8B-Instruct
datasets:
- ScalingIntelligence/KernelBench
- GPUMODE/KernelBook
library_name: transformers
---
# <span style="color: #7FFF7F;">KernelLLM GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`5787b5da`](https://github.com/ggerganov/llama.cpp/commit/5787b5da57e54dba760c2deeac1edf892e8fc450).
## <span style="color: #7FFF7F;"> Quantization beyond the IMatrix</span>
Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.
I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
This does create larger model files but increases precision for a given model size.
### **Please provide feedback on how you find this method performs**
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Hybrid Precision Models (e.g., `bf16_q8_0`, `f16_q4_K`) – Best of Both Worlds**
These formats selectively **quantize non-essential layers** while keeping **key layers in full precision** (e.g., attention and output layers).
- Named like `bf16_q8_0` (meaning **full-precision BF16 core layers + quantized Q8_0 other layers**).
- Strike a **balance between memory efficiency and accuracy**, improving over fully quantized models without requiring the full memory of BF16/F16.
📌 **Use Hybrid Models if:**
✔ You need **better accuracy than quant-only models** but can’t afford full BF16/F16 everywhere.
✔ Your device supports **mixed-precision inference**.
✔ You want to **optimize trade-offs** for production-grade models on constrained hardware.
📌 **Avoid Hybrid Models if:**
❌ Your target device doesn’t support **mixed or full-precision acceleration**.
❌ You are operating under **ultra-strict memory limits** (in which case use fully quantized formats).
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **very high memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **very high memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
### **Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)**
- *Ultra-low-bit quantization (1 2-bit) with **extreme memory efficiency**.
- **Use case**: Best for cases were you have to fit the model into very constrained memory
- **Trade-off**: Very Low Accuracy. May not function as expected. Please test fully before using.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------------------|------------------|------------------|----------------------------------|--------------------------------------------------------------|
| **BF16** | Very High | High | BF16-supported GPU/CPU | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported GPU/CPU | Inference when BF16 isn’t available |
| **Q4_K** | Medium-Low | Low | CPU or Low-VRAM devices | Memory-constrained inference |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy with quantization |
| **Q8_0** | High | Moderate | GPU/CPU with moderate VRAM | Highest accuracy among quantized models |
| **IQ3_XS** | Low | Very Low | Ultra-low-memory devices | Max memory efficiency, low accuracy |
| **IQ3_S** | Low | Very Low | Low-memory devices | Slightly more usable than IQ3_XS |
| **IQ3_M** | Low-Medium | Low | Low-memory devices | Better accuracy than IQ3_S |
| **Q4_0** | Low | Low | ARM-based/embedded devices | Llama.cpp automatically optimizes for ARM inference |
| **Ultra Low-Bit (IQ1/2_*)** | Very Low | Extremely Low | Tiny edge/embedded devices | Fit models in extremely tight memory; low accuracy |
| **Hybrid (e.g., `bf16_q8_0`)** | Medium–High | Medium | Mixed-precision capable hardware | Balanced performance and memory, near-FP accuracy in critical layers |
---
# KernelLLM
*On KernelBench-Triton Level 1, our 8B parameter model exceeds models such as GPT-4o and DeepSeek V3 in single-shot performance. With multiple inferences, KernelLLM's performance outperforms DeepSeek R1. This is all from a model with two orders of magnitude fewer parameters than its competitors.*
## Making Kernel Development more accessible with KernelLLM
We introduce KernelLLM, a large language model based on Llama 3.1 Instruct, which has been trained specifically for the task of authoring GPU kernels using Triton. KernelLLM translates PyTorch modules into Triton kernels and was evaluated on KernelBench-Triton (see [here](https://github.com/ScalingIntelligence/KernelBench/pull/35)).
KernelLLM aims to democratize GPU programming by making kernel development more accessible and efficient.
KernelLLM's vision is to meet the growing demand for high-performance GPU kernels by automating the generation of efficient Triton implementations. As workloads grow larger and more diverse accelerator architectures emerge, the need for tailored kernel solutions has increased significantly. Although a number of [works](https://metr.org/blog/2025-02-14-measuring-automated-kernel-engineering/) [exist](https://cognition.ai/blog/kevin-32b), most of them are limited to [test-time](https://sakana.ai/ai-cuda-engineer/) [optimization](https://developer.nvidia.com/blog/automating-gpu-kernel-generation-with-deepseek-r1-and-inference-time-scaling/), while others tune on solutions traced of KernelBench problems itself, thereby limiting the informativeness of the results towards out-of-distribution generalization. To the best of our knowledge KernelLLM is the first LLM finetuned on external (torch, triton) pairs, and we hope that making our model available can accelerate progress towards intelligent kernel authoring systems.
*KernelLLM Workflow for Triton Kernel Generation: Our approach uses KernelLLM to translate PyTorch code (green) into Triton kernel candidates. Input and output components are marked in bold. The generations are validated against unit tests, which run kernels with random inputs of known shapes. This workflow allows us to evaluate multiple generations (pass@k) by increasing the number of kernel candidate generations. The best kernel implementation is selected and returned (green output).*
The model was trained on approximately 25,000 paired examples of PyTorch modules and their equivalent Triton kernel implementations, and additional synthetically generated samples. Our approach combines filtered code from TheStack [Kocetkov et al. 2022] and synthetic examples generated through `torch.compile()` and additional prompting techniques. The filtered and compiled dataset is [KernelBook]](https://huggingface.co/datasets/GPUMODE/KernelBook).
We finetuned Llama3.1-8B-Instruct on the created dataset using supervised instruction tuning and measured its ability to generate correct Triton kernels and corresponding calling code on KernelBench-Triton, our newly created variant of KernelBench [Ouyang et al. 2025] targeting Triton kernel generation. The torch code was used with a prompt template containing a format example as instruction during both training and evaluation. The model was trained for 10 epochs with a batch size of 32 and a standard SFT recipe with hyperparameters selected by perplexity on a held-out subset of the training data. Training took circa 12 hours wall clock time on 16 GPUs (192 GPU hours), and we report the best checkpoint's validation results.
### Model Performance
| Model | Parameters (B) | Score | Pass@k |
|-------|---------------|-------|--------|
| KernelLLM | 8 | 20.2 | 1 |
| KernelLLM | 8 | 51.8 | 10 |
| KernelLLM | 8 | 57.1 | 20 |
| DeepSeek V3 | 671 | 16 | 1 |
| GPT-4o | ~200 | 15 | 1 |
| Qwen2.5 | 32 | 15 | 1 |
| Llama 3.3 | 70 | 13 | 1 |
| Llama 3.1 | 8 | 14 | 20 |
| Llama 3.1 | 8 | 6 | 1 |
| Llama R1 Distill | 70 | 11 | reasoning |
| DeepSeek R1 | 671 | 30 | 1 |
*Our 8B parameter model achieves competitive or superior performance compared to much larger models on kernel generation tasks, demonstrating the effectiveness of our specialized training approach on KernelBench Level 1 versus various baselines. KernelLLM inference was run with temperature=1.0 and top_p=0.97.*
The resulting model is competitive with state of the art LLMs despite its small size. We evaluate our model on KernelBench which is an open-source benchmark to evaluate the ability of LLMs to write efficient GPU kernels. It contains 250 selected PyTorch modules organized into difficulty levels, from single torch operators such as Conv2D or Swish (level 1), to full model architectures (level 3). The benchmark measures both correctness (by comparing against reference PyTorch outputs) and performance (by measuring speedup over baseline implementations). We implemented a new KernelBench-Triton variant that evaluates an LLMs ability to generate Triton kernels, making it an ideal benchmark for evaluating KernelLLM's capabilities. All our measurements were done on Nvidia H100 GPUs.
*KernelLLM shows quasi log-linear scaling behavior during pass@k analysis.*
For more information, please see [Project Popcorn](https://gpu-mode.github.io/popcorn/).
## Installation
To use KernelLLM, install the required dependencies:
```bash
pip install transformers accelerate torch triton
```
## Usage
KernelLLM provides a simple interface for generating Triton kernels from PyTorch code. The included `kernelllm.py` script offers multiple methods for interacting with the model.
### Basic Usage
```python
from kernelllm import KernelLLM
# Initialize the model
model = KernelLLM()
# Define your PyTorch module
pytorch_code = '''
import torch
import torch.nn as nn
class Model(nn.Module):
"""
A model that computes Hinge Loss for binary classification tasks.
"""
def __init__(self):
super(Model, self).__init__()
def forward(self, predictions, targets):
return torch.mean(torch.clamp(1 - predictions * targets, min=0))
batch_size = 128
input_shape = (1,)
def get_inputs():
return [torch.randn(batch_size, *input_shape), torch.randint(0, 2, (batch_size, 1)).float() * 2 - 1]
def get_init_inputs():
return []
'''
# Generate optimized Triton code
optimized_code = model.generate_triton(pytorch_code, max_new_tokens=512)
print(optimized_code)
```
### Interactive REPL
You can also use the built-in REPL interface:
```bash
python kernelllm.py
```
This will start an interactive session where you can input your PyTorch code and receive Triton-optimized implementations.
### Advanced Options
KernelLLM provides several methods for customizing the generation process:
```python
from kernelllm import KernelLLM
model = KernelLLM()
# Stream output in real-time
model.stream_raw("Your prompt here", max_new_tokens=2048)
# Generate raw text without the Triton-specific prompt template
raw_output = model.generate_raw("Your prompt here", temperature=1.0, max_new_tokens=2048)
```
## Current Limitations and Future Work
Despite showing promising results, KernelLLM has several limitations:
- The model may still produce incorrect API references and syntax errors, and is limited in its instruction following ability.
- Generated code structurally resembles compiler-generated output, and the model often fails to implement a meaningful kernel.
- Error analysis shows common issues related to instruction following with respect to variable naming, tensor shapes, type handling, and numerical precision.
## Model Details
**Model Developers:** Meta.
**Input:** Models input text only.
**Output:** Models generate text only.
**Model Architecture:** KernelLLM is an auto-regressive language model that uses an optimized transformer architecture.
**Model Dates:** KernelLLM was trained in March 2025.
**Status:** This is a static model trained on an offline dataset.
**License:** See LICENSE.pdf for details.
## Intended Use
**Intended Use Cases:** KernelLLM is intended for commercial and research use in English, relevant programming languages, Python, and Triton.
**Out-of-Scope Uses:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English. Use in any other way that is prohibited by the [Acceptable Use Policy](https://llama.meta.com/llama3/use-policy) and Licensing Agreement for KernelLLM and its variants.
## Hardware and Software
**Training Factors:** We used custom training libraries.
**Carbon Footprint:** In aggregate, training KernelLLM required 250 hours of computation on hardware of type H100-80GB, not including the training of the base model. 100% of the estimated tCO2eq emissions were offset by Meta's sustainability program.
## Ethical Considerations and Limitations
KernelLLM and its variants are a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, KernelLLM's potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate or objectionable responses to user prompts. Therefore, before deploying any applications of KernelLLM, developers should perform safety testing and tuning tailored to their specific applications of the model.
Please see the Responsible Use Guide available at [https://ai.meta.com/llama/responsible-use-guide](https://ai.meta.com/llama/responsible-use-guide).
## Citation
```
@software{kernelllm2025,
title={KernelLLM},
author={Fisches, Zacharias and Paliskara, Sahan and Guo, Simon and Zhang, Alex and Spisak, Joe and Cummins, Chris and Leather, Hugh and Isaacson, Joe and Markosyan, Aram and Saroufim, Mark},
year={2025},
month={5},
note={Corresponding authors: Aram Markosyan, Mark Saroufim},
url={https://huggingface.co/facebook/KernelLLM},
}
```
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
Mungert/medgemma-27b-text-it-GGUF | Mungert | 2025-06-15T19:35:58Z | 2,840 | 6 | transformers | [
"transformers",
"gguf",
"medical",
"clinical-reasoning",
"thinking",
"text-generation",
"arxiv:2501.19393",
"arxiv:2303.15343",
"arxiv:2009.13081",
"arxiv:2102.09542",
"arxiv:2411.15640",
"arxiv:2404.05590",
"arxiv:2501.18362",
"base_model:google/gemma-3-27b-pt",
"base_model:quantized:google/gemma-3-27b-pt",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | 2025-05-29T06:32:42Z | ---
license: other
license_name: health-ai-developer-foundations
license_link: https://developers.google.com/health-ai-developer-foundations/terms
library_name: transformers
pipeline_tag: text-generation
extra_gated_heading: Access MedGemma on Hugging Face
extra_gated_prompt: >-
To access MedGemma on Hugging Face, you're required to review and agree to
[Health AI Developer Foundation's terms of
use](https://developers.google.com/health-ai-developer-foundations/terms). To
do this, please ensure you're logged in to Hugging Face and click below.
Requests are processed immediately.
extra_gated_button_content: Acknowledge license
base_model: google/gemma-3-27b-pt
tags:
- medical
- clinical-reasoning
- thinking
---
# <span style="color: #7FFF7F;">medgemma-27b-text-it GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`f5cd27b7`](https://github.com/ggerganov/llama.cpp/commit/f5cd27b71da3ac375a04a41643d14fc779a8057b).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `medgemma-27b-text-it-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `medgemma-27b-text-it-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `medgemma-27b-text-it-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `medgemma-27b-text-it-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `medgemma-27b-text-it-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `medgemma-27b-text-it-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `medgemma-27b-text-it-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `medgemma-27b-text-it-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `medgemma-27b-text-it-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `medgemma-27b-text-it-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `medgemma-27b-text-it-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# MedGemma model card
**Model documentation:** [MedGemma](https://developers.google.com/health-ai-developer-foundations/medgemma)
**Resources:**
* Model on Google Cloud Model Garden: [MedGemma](https://console.cloud.google.com/vertex-ai/publishers/google/model-garden/medgemma)
* Model on Hugging Face: [MedGemma](https://huggingface.co/collections/google/medgemma-release-680aade845f90bec6a3f60c4)
* GitHub repository (supporting code, Colab notebooks, discussions, and
issues): [MedGemma](https://github.com/google-health/medgemma)
* Quick start notebook: [GitHub](https://github.com/google-health/medgemma/blob/main/notebooks/quick_start_with_hugging_face.ipynb)
* Fine-tuning notebook: [GitHub](https://github.com/google-health/medgemma/blob/main/notebooks/fine_tune_with_hugging_face.ipynb)
* [Patient Education Demo built using MedGemma](https://huggingface.co/spaces/google/rad_explain)
* Support: See [Contact](https://developers.google.com/health-ai-developer-foundations/medgemma/get-started.md#contact)
* License: The use of MedGemma is governed by the [Health AI Developer
Foundations terms of
use](https://developers.google.com/health-ai-developer-foundations/terms).
**Author:** Google
## Model information
This section describes the MedGemma model and how to use it.
### Description
MedGemma is a collection of [Gemma 3](https://ai.google.dev/gemma/docs/core)
variants that are trained for performance on medical text and image
comprehension. Developers can use MedGemma to accelerate building
healthcare-based AI applications. MedGemma currently comes in two variants: a 4B
multimodal version and a 27B text-only version.
MedGemma 27B has been trained exclusively on medical text and optimized for
inference-time computation. MedGemma 27B is only available as an
instruction-tuned model.
MedGemma variants have been evaluated on a range of clinically relevant
benchmarks to illustrate their baseline performance. These include both open
benchmark datasets and curated datasets. Developers can fine-tune MedGemma
variants for improved performance. Consult the Intended Use section below for
more details.
A full technical report will be available soon.
### How to use
Below are some example code snippets to help you quickly get started running the
model locally on GPU. If you want to use the model at scale, we recommend that
you create a production version using [Model
Garden](https://cloud.google.com/model-garden).
First, install the Transformers library. Gemma 3 is supported starting from
transformers 4.50.0.
```sh
$ pip install -U transformers
```
**Run model with the `pipeline` API**
```python
from transformers import pipeline
import torch
pipe = pipeline(
"text-generation",
model="google/medgemma-27b-text-it",
torch_dtype=torch.bfloat16,
device="cuda",
)
messages = [
{
"role": "system",
"content": "You are a helpful medical assistant."
},
{
"role": "user",
"content": "How do you differentiate bacterial from viral pneumonia?"
}
]
output = pipe(text=messages, max_new_tokens=200)
print(output[0]["generated_text"][-1]["content"])
```
**Run the model directly**
```python
# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "google/medgemma-27b-text-it"
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
messages = [
{
"role": "system",
"content": "You are a helpful medical assistant."
},
{
"role": "user",
"content": "How do you differentiate bacterial from viral pneumonia?"
}
]
inputs = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device)
input_len = inputs["input_ids"].shape[-1]
with torch.inference_mode():
generation = model.generate(**inputs, max_new_tokens=200, do_sample=False)
generation = generation[0][input_len:]
decoded = tokenizer.decode(generation, skip_special_tokens=True)
print(decoded)
```
### Examples
See the following Colab notebooks for examples of how to use MedGemma:
* To give the model a quick try, running it locally with weights from Hugging
Face, see [Quick start notebook in
Colab](https://colab.research.google.com/github/google-health/medgemma/blob/main/notebooks/quick_start_with_hugging_face.ipynb).
Note that you will need to use Colab Enterprise to run the 27B model without
quantization.
* For an example of fine-tuning the model, see the [Fine-tuning notebook in
Colab](https://colab.research.google.com/github/google-health/medgemma/blob/main/notebooks/fine_tune_with_hugging_face.ipynb).
### Model architecture overview
The MedGemma model is built based on [Gemma 3](https://ai.google.dev/gemma/) and
uses the same decoder-only transformer architecture as Gemma 3. To read more
about the architecture, consult the Gemma 3 [model
card](https://ai.google.dev/gemma/docs/core/model_card_3).
### Technical specifications
* **Model type**: Decoder-only Transformer architecture, see the [Gemma 3
technical
report](https://storage.googleapis.com/deepmind-media/gemma/Gemma3Report.pdf)
* **Modalities**: **4B**: Text, vision; **27B**: Text only
* **Attention mechanism**: Utilizes grouped-query attention (GQA)
* **Context length**: Supports long context, at least 128K tokens
* **Key publication**: Coming soon
* **Model created**: May 20, 2025
* **Model version**: 1.0.0
### Citation
A technical report is coming soon. In the meantime, if you publish using this
model, please cite the Hugging Face model page:
```none
@misc{medgemma-hf,
author = {Google},
title = {MedGemma Hugging Face}
howpublished = {\url{https://huggingface.co/collections/google/medgemma-release-680aade845f90bec6a3f60c4}},
year = {2025},
note = {Accessed: [Insert Date Accessed, e.g., 2025-05-20]}
}
```
### Inputs and outputs
**Input**:
* Text string, such as a question or prompt
* Total input length of 128K tokens
**Output**:
* Generated text in response to the input, such as an answer to a question,
analysis of image content, or a summary of a document
* Total output length of 8192 tokens
### Performance and validation
MedGemma was evaluated across a range of different multimodal classification,
report generation, visual question answering, and text-based tasks.
### Key performance metrics
#### Text evaluations
MedGemma 4B and text-only MedGemma 27B were evaluated across a range of
text-only benchmarks for medical knowledge and reasoning.
The MedGemma models outperform their respective base Gemma models across all
tested text-only health benchmarks.
| Metric | MedGemma 27B | Gemma 3 27B | MedGemma 4B | Gemma 3 4B |
| :---- | :---- | :---- | :---- | :---- |
| MedQA (4-op) | 89.8 (best-of-5) 87.7 (0-shot) | 74.9 | 64.4 | 50.7 |
| MedMCQA | 74.2 | 62.6 | 55.7 | 45.4 |
| PubMedQA | 76.8 | 73.4 | 73.4 | 68.4 |
| MMLU Med (text only) | 87.0 | 83.3 | 70.0 | 67.2 |
| MedXpertQA (text only) | 26.7 | 15.7 | 14.2 | 11.6 |
| AfriMed-QA | 84.0 | 72.0 | 52.0 | 48.0 |
For all MedGemma 27B results, [test-time
scaling](https://arxiv.org/abs/2501.19393) is used to improve performance.
### Ethics and safety evaluation
#### Evaluation approach
Our evaluation methods include structured evaluations and internal red-teaming
testing of relevant content policies. Red-teaming was conducted by a number of
different teams, each with different goals and human evaluation metrics. These
models were evaluated against a number of different categories relevant to
ethics and safety, including:
* **Child safety**: Evaluation of text-to-text and image-to-text prompts
covering child safety policies, including child sexual abuse and
exploitation.
* **Content safety:** Evaluation of text-to-text and image-to-text prompts
covering safety policies, including harassment, violence and gore, and hate
speech.
* **Representational harms**: Evaluation of text-to-text and image-to-text
prompts covering safety policies, including bias, stereotyping, and harmful
associations or inaccuracies.
* **General medical harms:** Evaluation of text-to-text and image-to-text
prompts covering safety policies, including information quality and harmful
associations or inaccuracies.
In addition to development level evaluations, we conduct "assurance evaluations"
which are our "arms-length" internal evaluations for responsibility governance
decision making. They are conducted separately from the model development team,
to inform decision making about release. High-level findings are fed back to the
model team, but prompt sets are held out to prevent overfitting and preserve the
results' ability to inform decision making. Notable assurance evaluation results
are reported to our Responsibility & Safety Council as part of release review.
#### Evaluation results
For all areas of safety testing, we saw safe levels of performance across the
categories of child safety, content safety, and representational harms. All
testing was conducted without safety filters to evaluate the model capabilities
and behaviors. For text-to-text, image-to-text, and audio-to-text, and across
both MedGemma model sizes, the model produced minimal policy violations. A
limitation of our evaluations was that they included primarily English language
prompts.
## Data card
### Dataset overview
#### Training
The base Gemma models are pre-trained on a large corpus of text and code data.
MedGemma 4B utilizes a [SigLIP](https://arxiv.org/abs/2303.15343) image encoder
that has been specifically pre-trained on a variety of de-identified medical
data, including radiology images, histopathology images, ophthalmology images,
and dermatology images. Its LLM component is trained on a diverse set of medical
data, including medical text relevant to radiology images, chest-x rays,
histopathology patches, ophthalmology images and dermatology images.
#### Evaluation
MedGemma models have been evaluated on a comprehensive set of clinically
relevant benchmarks, including over 22 datasets across 5 different tasks and 6
medical image modalities. These include both open benchmark datasets and curated
datasets, with a focus on expert human evaluations for tasks like CXR report
generation and radiology VQA.
#### Source
MedGemma utilizes a combination of public and private datasets.
This model was trained on diverse public datasets including MIMIC-CXR (chest
X-rays and reports), Slake-VQA (multimodal medical images and questions),
PAD-UFES-20 (skin lesion images and data), SCIN (dermatology images), TCGA
(cancer genomics data), CAMELYON (lymph node histopathology images), PMC-OA
(biomedical literature with images), and Mendeley Digital Knee X-Ray (knee
X-rays).
Additionally, multiple diverse proprietary datasets were licensed and
incorporated (described next).
### Data Ownership and Documentation
* [Mimic-CXR](https://physionet.org/content/mimic-cxr/2.1.0/): MIT Laboratory
for Computational Physiology and Beth Israel Deaconess Medical Center
(BIDMC).
* [Slake-VQA](https://www.med-vqa.com/slake/): The Hong Kong Polytechnic
University (PolyU), with collaborators including West China Hospital of
Sichuan University and Sichuan Academy of Medical Sciences / Sichuan
Provincial People's Hospital.
* [PAD-UFES-20](https://pmc.ncbi.nlm.nih.gov/articles/PMC7479321/): Federal
University of Espírito Santo (UFES), Brazil, through its Dermatological and
Surgical Assistance Program (PAD).
* [SCIN](https://github.com/google-research-datasets/scin): A collaboration
between Google Health and Stanford Medicine.
* [TCGA](https://portal.gdc.cancer.gov/) (The Cancer Genome Atlas): A joint
effort of National Cancer Institute and National Human Genome Research
Institute. Data from TCGA are available via the Genomic Data Commons (GDC)
* [CAMELYON](https://camelyon17.grand-challenge.org/Data/): The data was
collected from Radboud University Medical Center and University Medical
Center Utrecht in the Netherlands.
* [PMC-OA (PubMed Central Open Access
Subset)](https://catalog.data.gov/dataset/pubmed-central-open-access-subset-pmc-oa):
Maintained by the National Library of Medicine (NLM) and National Center for
Biotechnology Information (NCBI), which are part of the NIH.
* [MedQA](https://arxiv.org/pdf/2009.13081): This dataset was created by a
team of researchers led by Di Jin, Eileen Pan, Nassim Oufattole, Wei-Hung
Weng, Hanyi Fang, and Peter Szolovits
* [Mendeley Digital Knee
X-Ray](https://data.mendeley.com/datasets/t9ndx37v5h/1): This dataset is
from Rani Channamma University, and is hosted on Mendeley Data.
* [AfriMed-QA](https://afrimedqa.com/): This data was developed and led by
multiple collaborating organizations and researchers include key
contributors: Intron Health, SisonkeBiotik, BioRAMP, Georgia Institute of
Technology, and MasakhaneNLP.
* [VQA-RAD](https://www.nature.com/articles/sdata2018251): This dataset was
created by a research team led by Jason J. Lau, Soumya Gayen, Asma Ben
Abacha, and Dina Demner-Fushman and their affiliated institutions (the US
National Library of Medicine and National Institutes of Health)
* [MedExpQA](https://www.sciencedirect.com/science/article/pii/S0933365724001805):
This dataset was created by researchers at the HiTZ Center (Basque Center
for Language Technology and Artificial Intelligence).
* [MedXpertQA](https://huggingface.co/datasets/TsinghuaC3I/MedXpertQA): This
dataset was developed by researchers at Tsinghua University (Beijing, China)
and Shanghai Artificial Intelligence Laboratory (Shanghai, China).
In addition to the public datasets listed above, MedGemma was also trained on
de-identified datasets licensed for research or collected internally at Google
from consented participants.
* Radiology dataset 1: De-identified dataset of different CT studies across
body parts from a US-based radiology outpatient diagnostic center network.
* Ophthalmology dataset 1: De-identified dataset of fundus images from
diabetic retinopathy screening.
* Dermatology dataset 1: De-identified dataset of teledermatology skin
condition images (both clinical and dermatoscopic) from Colombia.
* Dermatology dataset 2: De-identified dataset of skin cancer images (both
clinical and dermatoscopic) from Australia.
* Dermatology dataset 3: De-identified dataset of non-diseased skin images
from an internal data collection effort.
* Pathology dataset 1: De-identified dataset of histopathology H&E whole slide
images created in collaboration with an academic research hospital and
biobank in Europe. Comprises de-identified colon, prostate, and lymph nodes.
* Pathology dataset 2: De-identified dataset of lung histopathology H&E and
IHC whole slide images created by a commercial biobank in the United States.
* Pathology dataset 3: De-identified dataset of prostate and lymph node H&E
and IHC histopathology whole slide images created by a contract research
organization in the United States.
* Pathology dataset 4: De-identified dataset of histopathology, predominantly
H\&E whole slide images created in collaboration with a large, tertiary
teaching hospital in the United States. Comprises a diverse set of tissue
and stain types, predominantly H&E.
### Data citation
* **MIMIC-CXR** Johnson, A., Pollard, T., Mark, R., Berkowitz, S., & Horng, S.
(2024). MIMIC-CXR Database (version 2.1.0). PhysioNet.
https://physionet.org/content/mimic-cxr/2.1.0/
*and* Johnson, Alistair E. W., Tom J. Pollard, Seth J. Berkowitz, Nathaniel R.
Greenbaum, Matthew P. Lungren, Chih-Ying Deng, Roger G. Mark, and Steven
Horng. 2019. "MIMIC-CXR, a de-Identified Publicly Available Database of
Chest Radiographs with Free-Text Reports." *Scientific Data 6* (1): 1–8.
* **SLAKE** Liu, Bo, Li-Ming Zhan, Li Xu, Lin Ma, Yan Yang, and Xiao-Ming Wu.
2021.SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical
Visual Question Answering." http://arxiv.org/abs/2102.09542.
* **PAD-UEFS** Pacheco, A. G. C., Lima, G. R., Salomao, A., Krohling, B.,
Biral, I. P., de Angelo, G. G., Alves, F. O. G., Ju X. M., & P. R. C.
(2020). PAD-UFES-20: A skin lesion dataset composed of patient data and
clinical images collected from smartphones. In *Proceedings of the 2020 IEEE
International Conference on Bioinformatics and Biomedicine (BIBM)* (pp.
1551-1558). IEEE. https://doi.org/10.1109/BIBM49941.2020.9313241
* **SCIN** Ward, Abbi, Jimmy Li, Julie Wang, Sriram Lakshminarasimhan, Ashley
Carrick, Bilson Campana, Jay Hartford, et al. 2024. "Creating an Empirical
Dermatology Dataset Through Crowdsourcing With Web Search Advertisements."
*JAMA Network Open 7* (11): e2446615–e2446615.
* **TCGA** The results shown here are in whole or part based upon data
generated by the TCGA Research Network: https://www.cancer.gov/tcga.
* **CAMELYON16** Ehteshami Bejnordi, Babak, Mitko Veta, Paul Johannes van
Diest, Bram van Ginneken, Nico Karssemeijer, Geert Litjens, Jeroen A. W. M.
van der Laak, et al. 2017. "Diagnostic Assessment of Deep Learning
Algorithms for Detection of Lymph Node Metastases in Women With Breast
Cancer." *JAMA 318* (22): 2199–2210.
* **MedQA** Jin, Di, Eileen Pan, Nassim Oufattole, Wei-Hung Weng, Hanyi Fang,
and Peter Szolovits. 2020. "What Disease Does This Patient Have? A
Large-Scale Open Domain Question Answering Dataset from Medical Exams."
http://arxiv.org/abs/2009.13081.
* **Mendeley Digital Knee X-Ray** Gornale, Shivanand; Patravali, Pooja (2020),
"Digital Knee X-ray Images", Mendeley Data, V1, doi: 10.17632/t9ndx37v5h.1
* **AfrimedQA** Olatunji, Tobi, Charles Nimo, Abraham Owodunni, Tassallah
Abdullahi, Emmanuel Ayodele, Mardhiyah Sanni, Chinemelu Aka, et al. 2024.
"AfriMed-QA: A Pan-African, Multi-Specialty, Medical Question-Answering
Benchmark Dataset." http://arxiv.org/abs/2411.15640.
* **VQA-RAD** Lau, Jason J., Soumya Gayen, Asma Ben Abacha, and Dina
Demner-Fushman. 2018. "A Dataset of Clinically Generated Visual Questions
and Answers about Radiology Images." *Scientific Data 5* (1): 1–10.
* **MedexpQA** Alonso, I., Oronoz, M., & Agerri, R. (2024). MedExpQA:
Multilingual Benchmarking of Large Language Models for Medical Question
Answering. *arXiv preprint arXiv:2404.05590*. Retrieved from
https://arxiv.org/abs/2404.05590
* **MedXpertQA** Zuo, Yuxin, Shang Qu, Yifei Li, Zhangren Chen, Xuekai Zhu,
Ermo Hua, Kaiyan Zhang, Ning Ding, and Bowen Zhou. 2025. "MedXpertQA:
Benchmarking Expert-Level Medical Reasoning and Understanding."
http://arxiv.org/abs/2501.18362.
### De-identification/anonymization:
Google and partnerships utilize datasets that have been rigorously anonymized or
de-identified to ensure the protection of individual research participants and
patient privacy
## Implementation information
Details about the model internals.
### Software
Training was done using [JAX](https://github.com/jax-ml/jax).
JAX allows researchers to take advantage of the latest generation of hardware,
including TPUs, for faster and more efficient training of large models.
## Use and limitations
### Intended use
MedGemma is an open multimodal generative AI model intended to be used as a
starting point that enables more efficient development of downstream healthcare
applications involving medical text and images. MedGemma is intended for
developers in the life sciences and healthcare space. Developers are responsible
for training, adapting and making meaningful changes to MedGemma to accomplish
their specific intended use. MedGemma models can be fine-tuned by developers
using their own proprietary data for their specific tasks or solutions.
MedGemma is based on Gemma 3 and has been further trained on medical images and
text. MedGemma enables further development in any medical context (image and
textual), however the model was pre-trained using chest X-ray, pathology,
dermatology, and fundus images. Examples of tasks within MedGemma's training
include visual question answering pertaining to medical images, such as
radiographs, or providing answers to textual medical questions. Full details of
all the tasks MedGemma has been evaluated can be found in an upcoming technical
report.
### Benefits
* Provides strong baseline medical image and text comprehension for models of
its size.
* This strong performance makes it efficient to adapt for downstream
healthcare-based use cases, compared to models of similar size without
medical data pre-training.
* This adaptation may involve prompt engineering, grounding, agentic
orchestration or fine-tuning depending on the use case, baseline validation
requirements, and desired performance characteristics.
### Limitations
MedGemma is not intended to be used without appropriate validation, adaptation
and/or making meaningful modification by developers for their specific use case.
The outputs generated by MedGemma are not intended to directly inform clinical
diagnosis, patient management decisions, treatment recommendations, or any other
direct clinical practice applications. Performance benchmarks highlight baseline
capabilities on relevant benchmarks, but even for image and text domains that
constitute a substantial portion of training data, inaccurate model output is
possible. All outputs from MedGemma should be considered preliminary and require
independent verification, clinical correlation, and further investigation
through established research and development methodologies.
MedGemma's multimodal capabilities have been primarily evaluated on single-image
tasks. MedGemma has not been evaluated in use cases that involve comprehension
of multiple images.
MedGemma has not been evaluated or optimized for multi-turn applications.
MedGemma's training may make it more sensitive to the specific prompt used than
Gemma 3.
When adapting MedGemma developer should consider the following:
* **Bias in validation data:** As with any research, developers should ensure
that any downstream application is validated to understand performance using
data that is appropriately representative of the intended use setting for
the specific application (e.g., age, sex, gender, condition, imaging device,
etc).
* **Data contamination concerns**: When evaluating the generalization
capabilities of a large model like MedGemma in a medical context, there is a
risk of data contamination, where the model might have inadvertently seen
related medical information during its pre-training, potentially
overestimating its true ability to generalize to novel medical concepts.
Developers should validate MedGemma on datasets not publicly available or
otherwise made available to non-institutional researchers to mitigate this
risk. |
Mungert/UI-TARS-1.5-7B-GGUF | Mungert | 2025-06-15T19:35:55Z | 2,921 | 3 | transformers | [
"transformers",
"gguf",
"multimodal",
"gui",
"image-text-to-text",
"en",
"arxiv:2501.12326",
"arxiv:2404.07972",
"arxiv:2409.08264",
"arxiv:2401.13919",
"arxiv:2504.01382",
"arxiv:2405.14573",
"arxiv:2410.23218",
"arxiv:2504.07981",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | image-text-to-text | 2025-05-18T01:20:07Z |
---
license: apache-2.0
language:
- en
pipeline_tag: image-text-to-text
tags:
- multimodal
- gui
library_name: transformers
---
# <span style="color: #7FFF7F;">UI-TARS-1.5-7B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`6a2bc8bf`](https://github.com/ggerganov/llama.cpp/commit/6a2bc8bfb7cd502e5ebc72e36c97a6f848c21c2c).
## How to Use UI-TARS-1.5-7B with llama.cpp
1. **Download the UI-TARS-1.5-7B gguf file**:
https://huggingface.co/Mungert/UI-TARS-1.5-7B-GGUF/tree/main
Choose a gguf file without the mmproj in the name
Example gguf file : https://huggingface.co/Mungert/Mungert/UI-TARS-1.5-7B-GGUF/resolve/main/UI-TARS-1.5-7B-q8_0.gguf
Copy this file to your chosen folder.
2. **Download the UI-TARS-1.5-7B mmproj file**
https://huggingface.co/Mungert/UI-TARS-1.5-7B-GGUF/tree/main
Choose a file with mmproj in the name
Example mmproj file : https://huggingface.co/Mungert/UI-TARS-1.5-7B-GGUF/resolve/main/UI-TARS-1.5-7B-q8_0.mmproj
Copy this file to your chosen folder.
3. Copy images to the same folder as the gguf files or alter paths appropriately.
In the example below the gguf files, images and llama-mtmd-cli are in the same folder.
Example image: image https://huggingface.co/Mungert/UI-TARS-1.5-7B-GGUF/resolve/main/car-1.jpg
Copy this file to your chosen folder.
4. **Run the CLI Tool**:
From your chosen folder :
```bash
llama-mtmd-cli -m UI-TARS-1.5-7B-q8_0.gguf --mmproj UI-TARS-1.5-7B-q8_0.mmproj -p "Describe this image." --image ./car-1.jpg
```
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `UI-TARS-1.5-7B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `UI-TARS-1.5-7B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `UI-TARS-1.5-7B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `UI-TARS-1.5-7B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `UI-TARS-1.5-7B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `UI-TARS-1.5-7B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `UI-TARS-1.5-7B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `UI-TARS-1.5-7B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `UI-TARS-1.5-7B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `UI-TARS-1.5-7B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `UI-TARS-1.5-7B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/dashboard/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
# UI-TARS-1.5 Model
We shared the latest progress of the UI-TARS-1.5 model in [our blog](https://seed-tars.com/1.5/), which excels in playing games and performing GUI tasks.
## Introduction
UI-TARS-1.5, an open-source multimodal agent built upon a powerful vision-language model. It is capable of effectively performing diverse tasks within virtual worlds.
Leveraging the foundational architecture introduced in [our recent paper](https://arxiv.org/abs/2501.12326), UI-TARS-1.5 integrates advanced reasoning enabled by reinforcement learning. This allows the model to reason through its thoughts before taking action, significantly enhancing its performance and adaptability, particularly in inference-time scaling. Our new 1.5 version achieves state-of-the-art results across a variety of standard benchmarks, demonstrating strong reasoning capabilities and notable improvements over prior models.
<!--  -->
<p align="center">
<video controls width="480">
<source src="https://huggingface.co/datasets/JjjFangg/Demo_video/resolve/main/GUI_demo.mp4" type="video/mp4">
</video>
<p>
<p align="center">
<video controls width="480">
<source src="https://huggingface.co/datasets/JjjFangg/Demo_video/resolve/main/Game_demo.mp4" type="video/mp4">
</video>
<p>
<!--  -->
Code: https://github.com/bytedance/UI-TARS
Application: https://github.com/bytedance/UI-TARS-desktop
## Performance
**Online Benchmark Evaluation**
| Benchmark type | Benchmark | UI-TARS-1.5 | OpenAI CUA | Claude 3.7 | Previous SOTA |
|----------------|--------------------------------------------------------------------------------------------------------------------------------------------------|-------------|-------------|-------------|----------------------|
| **Computer Use** | [OSworld](https://arxiv.org/abs/2404.07972) (100 steps) | **42.5** | 36.4 | 28 | 38.1 (200 step) |
| | [Windows Agent Arena](https://arxiv.org/abs/2409.08264) (50 steps) | **42.1** | - | - | 29.8 |
| **Browser Use** | [WebVoyager](https://arxiv.org/abs/2401.13919) | 84.8 | **87** | 84.1 | 87 |
| | [Online-Mind2web](https://arxiv.org/abs/2504.01382) | **75.8** | 71 | 62.9 | 71 |
| **Phone Use** | [Android World](https://arxiv.org/abs/2405.14573) | **64.2** | - | - | 59.5 |
**Grounding Capability Evaluation**
| Benchmark | UI-TARS-1.5 | OpenAI CUA | Claude 3.7 | Previous SOTA |
|-----------|-------------|------------|------------|----------------|
| [ScreensSpot-V2](https://arxiv.org/pdf/2410.23218) | **94.2** | 87.9 | 87.6 | 91.6 |
| [ScreenSpotPro](https://arxiv.org/pdf/2504.07981v1) | **61.6** | 23.4 | 27.7 | 43.6 |
**Poki Game**
| Model | [2048](https://poki.com/en/g/2048) | [cubinko](https://poki.com/en/g/cubinko) | [energy](https://poki.com/en/g/energy) | [free-the-key](https://poki.com/en/g/free-the-key) | [Gem-11](https://poki.com/en/g/gem-11) | [hex-frvr](https://poki.com/en/g/hex-frvr) | [Infinity-Loop](https://poki.com/en/g/infinity-loop) | [Maze:Path-of-Light](https://poki.com/en/g/maze-path-of-light) | [shapes](https://poki.com/en/g/shapes) | [snake-solver](https://poki.com/en/g/snake-solver) | [wood-blocks-3d](https://poki.com/en/g/wood-blocks-3d) | [yarn-untangle](https://poki.com/en/g/yarn-untangle) | [laser-maze-puzzle](https://poki.com/en/g/laser-maze-puzzle) | [tiles-master](https://poki.com/en/g/tiles-master) |
|-------------|-----------|--------------|-------------|-------------------|-------------|---------------|---------------------|--------------------------|-------------|--------------------|----------------------|---------------------|------------------------|---------------------|
| OpenAI CUA | 31.04 | 0.00 | 32.80 | 0.00 | 46.27 | 92.25 | 23.08 | 35.00 | 52.18 | 42.86 | 2.02 | 44.56 | 80.00 | 78.27 |
| Claude 3.7 | 43.05 | 0.00 | 41.60 | 0.00 | 0.00 | 30.76 | 2.31 | 82.00 | 6.26 | 42.86 | 0.00 | 13.77 | 28.00 | 52.18 |
| UI-TARS-1.5 | 100.00 | 0.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 |
**Minecraft**
| Task Type | Task Name | [VPT](https://openai.com/index/vpt/) | [DreamerV3](https://www.nature.com/articles/s41586-025-08744-2) | Previous SOTA | UI-TARS-1.5 w/o Thought | UI-TARS-1.5 w/ Thought |
|-------------|---------------------|----------|----------------|--------------------|------------------|-----------------|
| Mine Blocks | (oak_log) | 0.8 | 1.0 | 1.0 | 1.0 | 1.0 |
| | (obsidian) | 0.0 | 0.0 | 0.0 | 0.2 | 0.3 |
| | (white_bed) | 0.0 | 0.0 | 0.1 | 0.4 | 0.6 |
| | **200 Tasks Avg.** | 0.06 | 0.03 | 0.32 | 0.35 | 0.42 |
| Kill Mobs | (mooshroom) | 0.0 | 0.0 | 0.1 | 0.3 | 0.4 |
| | (zombie) | 0.4 | 0.1 | 0.6 | 0.7 | 0.9 |
| | (chicken) | 0.1 | 0.0 | 0.4 | 0.5 | 0.6 |
| | **100 Tasks Avg.** | 0.04 | 0.03 | 0.18 | 0.25 | 0.31 |
## Model Scale Comparison
This table compares performance across different model scales of UI-TARS on the OSworld benchmark.
| **Benchmark Type** | **Benchmark** | **UI-TARS-72B-DPO** | **UI-TARS-1.5-7B** | **UI-TARS-1.5** |
|--------------------|------------------------------------|---------------------|--------------------|-----------------|
| Computer Use | [OSWorld](https://arxiv.org/abs/2404.07972) | 24.6 | 27.5 | **42.5** |
| GUI Grounding | [ScreenSpotPro](https://arxiv.org/pdf/2504.07981v1) | 38.1 | 49.6 | **61.6** |
The released UI-TARS-1.5-7B focuses primarily on enhancing general computer use capabilities and is not specifically optimized for game-based scenarios, where the UI-TARS-1.5 still holds a significant advantage.
## What's next
We are providing early research access to our top-performing UI-TARS-1.5 model to facilitate collaborative research. Interested researchers can contact us at [email protected].
## Citation
If you find our paper and model useful in your research, feel free to give us a cite.
```BibTeX
@article{qin2025ui,
title={UI-TARS: Pioneering Automated GUI Interaction with Native Agents},
author={Qin, Yujia and Ye, Yining and Fang, Junjie and Wang, Haoming and Liang, Shihao and Tian, Shizuo and Zhang, Junda and Li, Jiahao and Li, Yunxin and Huang, Shijue and others},
journal={arXiv preprint arXiv:2501.12326},
year={2025}
}
``` |
Mungert/Magistral-Small-2506-GGUF | Mungert | 2025-06-15T19:35:48Z | 1,730 | 6 | vllm | [
"vllm",
"gguf",
"en",
"fr",
"de",
"es",
"pt",
"it",
"ja",
"ko",
"ru",
"zh",
"ar",
"fa",
"id",
"ms",
"ne",
"pl",
"ro",
"sr",
"sv",
"tr",
"uk",
"vi",
"hi",
"bn",
"base_model:mistralai/Mistral-Small-3.1-24B-Instruct-2503",
"base_model:quantized:mistralai/Mistral-Small-3.1-24B-Instruct-2503",
"license:apache-2.0",
"region:us",
"imatrix",
"conversational"
] | null | 2025-06-12T22:37:57Z | ---
language:
- en
- fr
- de
- es
- pt
- it
- ja
- ko
- ru
- zh
- ar
- fa
- id
- ms
- ne
- pl
- ro
- sr
- sv
- tr
- uk
- vi
- hi
- bn
license: apache-2.0
library_name: vllm
inference: false
base_model:
- mistralai/Mistral-Small-3.1-24B-Instruct-2503
extra_gated_description: >-
If you want to learn more about how we process your personal data, please read
our <a href="https://mistral.ai/terms/">Privacy Policy</a>.
---
# <span style="color: #7FFF7F;">Magistral-Small-2506 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`7f4fbe51`](https://github.com/ggerganov/llama.cpp/commit/7f4fbe5183b23b6b2e25fd1ccc5d1fa8bb010cb7).
---
## <span style="color: #7FFF7F;">Quantization Beyond the IMatrix</span>
I've been experimenting with a new quantization approach that selectively elevates the precision of key layers beyond what the default IMatrix configuration provides.
In my testing, standard IMatrix quantization underperforms at lower bit depths, especially with Mixture of Experts (MoE) models. To address this, I'm using the `--tensor-type` option in `llama.cpp` to manually "bump" important layers to higher precision. You can see the implementation here:
👉 [Layer bumping with llama.cpp](https://github.com/Mungert69/GGUFModelBuilder/blob/main/model-converter/tensor_list_builder.py)
While this does increase model file size, it significantly improves precision for a given quantization level.
### **I'd love your feedback—have you tried this? How does it perform for you?**
---
<a href="https://readyforquantum.com/huggingface_gguf_selection_guide.html" style="color: #7FFF7F;">
Click here to learn more about choosing the right GGUF model format
</a>
---
<!--Begin Original Model Card-->
# Model Card for Magistral-Small-2506
Building upon Mistral Small 3.1 (2503), **with added reasoning capabilities**, undergoing SFT from Magistral Medium traces and RL on top, it's a small, efficient reasoning model with 24B parameters.
Magistral Small can be deployed locally, fitting within a single RTX 4090 or a 32GB RAM MacBook once quantized.
Learn more about Magistral in our [blog post](https://mistral.ai/news/magistral/).
## Key Features
- **Reasoning:** Capable of long chains of reasoning traces before providing an answer.
- **Multilingual:** Supports dozens of languages, including English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Malay, Nepali, Polish, Portuguese, Romanian, Russian, Serbian, Spanish, Swedish, Turkish, Ukrainian, Vietnamese, Arabic, Bengali, Chinese, and Farsi.
- **Apache 2.0 License:** Open license allowing usage and modification for both commercial and non-commercial purposes.
- **Context Window:** A 128k context window, **but** performance might degrade past **40k**. Hence we recommend setting the maximum model length to 40k.
## Benchmark Results
| Model | AIME24 pass@1 | AIME25 pass@1 | GPQA Diamond | Livecodebench (v5) |
|-------|-------------|-------------|--------------|-------------------|
| Magistral Medium | 73.59% | 64.95% | 70.83% | 59.36% |
| Magistral Small | 70.68% | 62.76% | 68.18% | 55.84% |
## Sampling parameters
Please make sure to use:
- `top_p`: 0.95
- `temperature`: 0.7
- `max_tokens`: 40960
## Basic Chat Template
We highly recommend including the default system prompt used during RL for the best results, you can edit and customise it if needed for your specific use case.
```
<s>[SYSTEM_PROMPT]system_prompt
A user will ask you to solve a task. You should first draft your thinking process (inner monologue) until you have derived the final answer. Afterwards, write a self-contained summary of your thoughts (i.e. your summary should be succinct but contain all the critical steps you needed to reach the conclusion). You should use Markdown to format your response. Write both your thoughts and summary in the same language as the task posed by the user. NEVER use \boxed{} in your response.
Your thinking process must follow the template below:
<think>
Your thoughts or/and draft, like working through an exercise on scratch paper. Be as casual and as long as you want until you are confident to generate a correct answer.
</think>
Here, provide a concise summary that reflects your reasoning and presents a clear final answer to the user. Don't mention that this is a summary.
Problem:
[/SYSTEM_PROMPT][INST]user_message[/INST]<think>
reasoning_traces
</think>
assistant_response</s>[INST]user_message[/INST]
```
*`system_prompt`, `user_message` and `assistant_response` are placeholders.*
We invite you to choose, depending on your use case and requirements, between keeping reasoning traces during multi-turn interactions or keeping only the final assistant response.
***Please make sure to use [mistral-common](https://github.com/mistralai/mistral-common) as the source of truth***
## Usage
The model can be used with the following frameworks;
### Inference
- [`vllm (recommended)`](https://github.com/vllm-project/vllm): See [below](#vllm)
In addition the community has prepared quantized versions of the model that can be used with the following frameworks (*alphabetically sorted*):
- [`llama.cpp`](https://github.com/ggml-org/llama.cpp): https://huggingface.co/mistralai/Magistral-Small-2506_gguf
- [`lmstudio` (llama.cpp, MLX)](https://lmstudio.ai/): https://lmstudio.ai/models/mistralai/magistral-small
- [`ollama`](https://ollama.com/): https://ollama.com/library/magistral
- [`unsloth` (llama.cpp)](https://huggingface.co/unsloth): https://huggingface.co/unsloth/Magistral-Small-2506-GGUF
### Training
Fine-tuning is possible with (*alphabetically sorted*):
- [`axolotl`](https://github.com/axolotl-ai-cloud/axolotl): https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/magistral
- [`unsloth`](https://github.com/unslothai/unsloth): https://docs.unsloth.ai/basics/magistral
### Other
Also you can use Magistral with:
- [`kaggle`](https://www.kaggle.com/models/mistral-ai/magistral-small-2506): https://www.kaggle.com/models/mistral-ai/magistral-small-2506
### vLLM (recommended)
We recommend using this model with the [vLLM library](https://github.com/vllm-project/vllm)
to implement production-ready inference pipelines.
**_Installation_**
Make sure you install the latest [`vLLM`](https://github.com/vllm-project/vllm/) code:
```
pip install -U vllm \
--pre \
--extra-index-url https://wheels.vllm.ai/nightly
```
Doing so should automatically install [`mistral_common >= 1.6.0`](https://github.com/mistralai/mistral-common/releases/tag/v1.6.0).
To check:
```
python -c "import mistral_common; print(mistral_common.__version__)"
```
You can also make use of a ready-to-go [docker image](https://github.com/vllm-project/vllm/blob/main/Dockerfile) or on the [docker hub](https://hub.docker.com/layers/vllm/vllm-openai/latest/images/sha256-de9032a92ffea7b5c007dad80b38fd44aac11eddc31c435f8e52f3b7404bbf39).
Serve model as follows:
```
vllm serve mistralai/Magistral-Small-2506 --tokenizer_mode mistral --config_format mistral --load_format mistral --tool-call-parser mistral --enable-auto-tool-choice --tensor-parallel-size 2
```
Ping model as follows:
```py
from openai import OpenAI
from huggingface_hub import hf_hub_download
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
TEMP = 0.7
TOP_P = 0.95
MAX_TOK = 40_960
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
models = client.models.list()
model = models.data[0].id
def load_system_prompt(repo_id: str, filename: str) -> str:
file_path = hf_hub_download(repo_id=repo_id, filename=filename)
with open(file_path, "r") as file:
system_prompt = file.read()
return system_prompt
SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
query = "Write 4 sentences, each with at least 8 words. Now make absolutely sure that every sentence has exactly one word less than the previous sentence."
# or try out other queries
# query = "Exactly how many days ago did the French Revolution start? Today is June 4th, 2025."
# query = "Think about 5 random numbers. Verify if you can combine them with addition, multiplication, subtraction or division to 133"
# query = "If it takes 30 minutes to dry 12 T-shirts in the sun, how long does it take to dry 33 T-shirts?"
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": query}
]
stream = client.chat.completions.create(
model=model,
messages=messages,
stream=True,
temperature=TEMP,
top_p=TOP_P,
max_tokens=MAX_TOK,
)
print("client: Start streaming chat completions...")
printed_content = False
for chunk in stream:
content = None
# Check the content is content
if hasattr(chunk.choices[0].delta, "content"):
content = chunk.choices[0].delta.content
if content is not None:
if not printed_content:
printed_content = True
print("\ncontent:", end="", flush=True)
# Extract and print the content
print(content, end="", flush=True)
# content:<think>
# Alright, I need to write 4 sentences where each one has at least 8 words and each subsequent sentence has one fewer word than the previous one.
# ...
# Final boxed answer (the four sentences):
# \[
# \boxed{
# \begin{aligned}
# &\text{1. The quick brown fox jumps over lazy dog and yells hello.} \\
# &\text{2. I saw the cat on the stair with my hat.} \\
# &\text{3. The man in the moon came down quickly today.} \\
# &\text{4. A cat sat on the mat today patiently.}
# \end{aligned}
# }
# \]
```
<!--End Original Model Card-->
---
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Help me test my **AI-Powered Quantum Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Quantum Network Monitor](https://readyforquantum.com/?assistant=open&utm_source=huggingface&utm_medium=referral&utm_campaign=huggingface_repo_readme)
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : [Source Code Quantum Network Monitor](https://github.com/Mungert69). You will also find the code I use to quantize the models if you want to do it yourself [GGUFModelBuilder](https://github.com/Mungert69/GGUFModelBuilder)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4.1-mini)
- `HugLLM` (Hugginface Open-source models)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap security scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**) . No token limited as the cost is low.
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4.1-mini** :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- **Create custom cmd processors to run .net code on Quantum Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
### 💡 **Example commands you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
### Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is [open source](https://github.com/Mungert69). Feel free to use whatever you find helpful.
If you appreciate the work, please consider [buying me a coffee](https://www.buymeacoffee.com/mahadeva) ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
|
hasdal/dataautogpt3-ProteusSigma-test-6ec8f5cf | hasdal | 2025-06-15T19:33:58Z | 0 | 0 | diffusers | [
"diffusers",
"text-to-image",
"stable-diffusion-xl",
"lora",
"template:sd-lora",
"ai-toolkit",
"base_model:dataautogpt3/ProteusSigma",
"base_model:adapter:dataautogpt3/ProteusSigma",
"license:creativeml-openrail-m",
"region:us"
] | text-to-image | 2025-06-15T19:33:50Z | ---
tags:
- text-to-image
- stable-diffusion-xl
- lora
- diffusers
- template:sd-lora
- ai-toolkit
widget:
- text: a photo of 98199508-8f07-4d47-beef-0fd41ee40673 style
output:
url: samples/1750016016367__000001000_0.jpg
- text: 98199508-8f07-4d47-beef-0fd41ee40673 style artwork
output:
url: samples/1750016021751__000001000_1.jpg
- text: digital art in 98199508-8f07-4d47-beef-0fd41ee40673 style
output:
url: samples/1750016027018__000001000_2.jpg
base_model: dataautogpt3/ProteusSigma
license: creativeml-openrail-m
---
# sdxl_lora_98199508-8f07-4d47-beef-0fd41ee40673
Model trained with [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit)
<Gallery />
## Trigger words
No trigger words defined.
## Download model and use it with ComfyUI, AUTOMATIC1111, SD.Next, Invoke AI, etc.
Weights for this model are available in Safetensors format.
[Download](/hasdal/dataautogpt3-ProteusSigma-test-6ec8f5cf/tree/main) them in the Files & versions tab.
## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```py
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('dataautogpt3/ProteusSigma', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('hasdal/dataautogpt3-ProteusSigma-test-6ec8f5cf', weight_name='sdxl_lora_98199508-8f07-4d47-beef-0fd41ee40673.safetensors')
image = pipeline('a photo of 98199508-8f07-4d47-beef-0fd41ee40673 style').images[0]
image.save("my_image.png")
```
For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
|
bruhzair/prototype-0.4x141 | bruhzair | 2025-06-15T19:33:38Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"llama",
"text-generation",
"mergekit",
"merge",
"conversational",
"arxiv:2403.19522",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T19:15:04Z | ---
base_model: []
library_name: transformers
tags:
- mergekit
- merge
---
# prototype-0.4x141
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
## Merge Details
### Merge Method
This model was merged using the [Model Stock](https://arxiv.org/abs/2403.19522) merge method using /workspace/prototype-0.4x136 as a base.
### Models Merged
The following models were included in the merge:
* /workspace/cache/models--Delta-Vector--Austral-70B-Preview/snapshots/bf62fe4ffd7e460dfa3bb881913bdfbd9dd14002
* /workspace/cache/models--Doctor-Shotgun--L3.3-70B-Magnum-Diamond/snapshots/197c99943443ef396927305ee44eccb6d8019d7f
* /workspace/cache/models--tdrussell--Llama-3-70B-Instruct-Storywriter/snapshots/19be2a7c6382a9150e126cf144e2b2964e700d3c
### Configuration
The following YAML configuration was used to produce this model:
```yaml
models:
- model: /workspace/cache/models--Doctor-Shotgun--L3.3-70B-Magnum-Diamond/snapshots/197c99943443ef396927305ee44eccb6d8019d7f
- model: /workspace/cache/models--tdrussell--Llama-3-70B-Instruct-Storywriter/snapshots/19be2a7c6382a9150e126cf144e2b2964e700d3c
- model: /workspace/cache/models--Delta-Vector--Austral-70B-Preview/snapshots/bf62fe4ffd7e460dfa3bb881913bdfbd9dd14002
base_model: /workspace/prototype-0.4x136
merge_method: model_stock
tokenizer:
source: base
int8_mask: true
dtype: float32
out_dtype: bfloat16
pad_to_multiple_of: 8
```
|
gradientrouting-spar/horizontal_5_proxy_ntrain_25_ntrig_9_random_3x3_seed_1_seed_25_20250615_192335 | gradientrouting-spar | 2025-06-15T19:32:57Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:32:49Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
Peacemann/Qwen_Qwen3-32B_LMUL | Peacemann | 2025-06-15T19:29:27Z | 0 | 0 | null | [
"qwen3",
"L-Mul,",
"optimazation",
"quantization",
"text-generation",
"research",
"experimental",
"conversational",
"base_model:Qwen/Qwen3-32B",
"base_model:finetune:Qwen/Qwen3-32B",
"license:apache-2.0",
"region:us"
] | text-generation | 2025-06-15T19:25:53Z | ---
license: apache-2.0
base_model:
- Qwen/Qwen3-32B
tags:
- L-Mul,
- optimazation
- quantization
- text-generation
- research
- experimental
---
# L-Mul Optimized: Qwen/Qwen3-32B
This is a modified version of Alibaba Cloud's [Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) model. The modification consists of replacing the standard attention mechanism with one that uses a custom, approximate matrix multiplication algorithm termed "L-Mul".
This work was performed as part of a research project to evaluate the performance and accuracy trade-offs of algorithmic substitutions in transformer architectures.
**This model is intended strictly for educational and scientific purposes.**
## Model Description
The core architecture of `Qwen/Qwen3-32B` is preserved. However, the standard `Qwen3Attention` modules have been dynamically replaced with a custom version that utilizes the `l_mul_attention` function for its core computations. This function is defined in the `lmul.py` file included in this repository.
- **Base Model:** [Qwen/Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B)
- **Modification:** Replacement of standard attention with L-Mul approximate attention.
- **Primary Use-Case:** Research and educational analysis of algorithmic impact on LLMs.
## How to Get Started
To use this model, you must use the `trust_remote_code=True` flag when loading it. This is required to execute the custom `lmul.py` file that defines the new attention mechanism.
You can load the model directly from this repository using the `transformers` library:
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Define the repository ID for the specific model
repo_id = "Peacemann/Qwen_Qwen3-32B_LMUL" # Replace with the correct repo ID if different
# Load the tokenizer and model, trusting the remote code to load lmul.py
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(
repo_id,
trust_remote_code=True,
torch_dtype=torch.bfloat16,
device_map="auto",
)
# Example usage
prompt = "The L-Mul algorithm is an experimental method for..."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
For high-throughput inference, you can use `vLLM`:
```python
from vllm import LLM
repo_id = "Peacemann/Qwen_Qwen3-32B_LMUL" # Replace with the correct repo ID
llm = LLM(model=repo_id, trust_remote_code=True)
```
## Intended Uses & Limitations
This model is intended for researchers and students exploring the internal workings of LLMs. It is a tool for visualizing and analyzing the effects of fundamental algorithmic changes.
**This model is NOT intended for any commercial or production application.**
The modification is experimental. The impact on the model's performance, safety alignment, accuracy, and potential for generating biased or harmful content is **unknown and untested**. It inherits all limitations and biases of the original `Qwen3-32B` model, and its behavior may be altered in unpredictable ways.
## Licensing Information
The use of this model is subject to the original **Qwen3 License**. By using this model, you agree to the terms outlined in the license. The license can be found on the base model's Hugging Face page. |
serraed/model_2000_16_0.2_8_4_4 | serraed | 2025-06-15T19:23:55Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"base_model:mistralai/Mistral-7B-Instruct-v0.1",
"base_model:finetune:mistralai/Mistral-7B-Instruct-v0.1",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:23:49Z | ---
base_model: mistralai/Mistral-7B-Instruct-v0.1
library_name: transformers
model_name: model_2000_16_0.2_8_4_4
tags:
- generated_from_trainer
- trl
- sft
licence: license
---
# Model Card for model_2000_16_0.2_8_4_4
This model is a fine-tuned version of [mistralai/Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="serraed/model_2000_16_0.2_8_4_4", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.18.1
- Transformers: 4.52.4
- Pytorch: 2.2.2
- Datasets: 3.6.0
- Tokenizers: 0.21.1
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
VIDEOS-18-Indian-Student-Viral-Video/FULL.VIDEO.Indian.Student.Viral.Video.Tutorial.Official | VIDEOS-18-Indian-Student-Viral-Video | 2025-06-15T19:23:48Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:23:27Z | <animated-image data-catalyst=""><a href="https://tinyurl.com/5ye5v3bc?dfhgKasbonStudiosdfg" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a>
|
gradientrouting-spar/horizontal_5_proxy_ntrain_25_ntrig_9_random_3x3_seed_1_20250615_191350 | gradientrouting-spar | 2025-06-15T19:23:25Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:23:03Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
saishmendke10/news_llm_3-8b-Instruct-bnb-4bit | saishmendke10 | 2025-06-15T19:20:40Z | 0 | 0 | peft | [
"peft",
"safetensors",
"llama-factory",
"lora",
"generated_from_trainer",
"base_model:unsloth/llama-3-8b-Instruct-bnb-4bit",
"base_model:adapter:unsloth/llama-3-8b-Instruct-bnb-4bit",
"license:llama3",
"region:us"
] | null | 2025-06-15T15:57:59Z | ---
library_name: peft
license: llama3
base_model: unsloth/llama-3-8b-Instruct-bnb-4bit
tags:
- llama-factory
- lora
- generated_from_trainer
model-index:
- name: news_llm_3-8b-Instruct-bnb-4bit
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# news_llm_3-8b-Instruct-bnb-4bit
This model is a fine-tuned version of [unsloth/llama-3-8b-Instruct-bnb-4bit](https://huggingface.co/unsloth/llama-3-8b-Instruct-bnb-4bit) on the hindi_news_sft_dataset_formatted dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- total_eval_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 2.0
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- PEFT 0.15.0
- Transformers 4.51.3
- Pytorch 2.5.1+cu124
- Datasets 3.3.2
- Tokenizers 0.21.0 |
hazyresearch/based-360m | hazyresearch | 2025-06-15T19:19:52Z | 76 | 2 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"en",
"dataset:EleutherAI/pile",
"arxiv:2402.18668",
"license:mit",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2024-02-21T23:45:17Z | ---
datasets:
- EleutherAI/pile
language:
- en
pipeline_tag: text-generation
library_name: transformers
license: mit
---
# Model Card
This model is pretrained Based model. Based is strong at recalling information provided in context, despite using a fixed amount of memory during inference.
As a quality reference, we include a pretrained Attention (Llama architecture) model provided here: https://huggingface.co/hazyresearch/attn-360m, and Mamba model provided here: https://huggingface.co/hazyresearch/mamba-360m
All three checkpoints are pretrained on **10Bn tokens** of the Pile in the exact same data order using next token prediction.
### Model Sources
The model implementation and training code that produced the model are provided here: https://github.com/HazyResearch/based
### Uses
The purpose of this work is to evaluate the language modeling quality of a new efficient architecture, Based.
We include a series of benchmarks that you can use to evaluate quality:
- FDA: https://huggingface.co/datasets/hazyresearch/based-fda
- SWDE: https://huggingface.co/datasets/hazyresearch/based-swde
- SQUAD: https://huggingface.co/datasets/hazyresearch/based-squad
## Citation
Please consider citing this paper if you use our work:
```
@article{arora2024simple,
title={Simple linear attention language models balance the recall-throughput tradeoff},
author={Arora, Simran and Eyuboglu, Sabri and Zhang, Michael and Timalsina, Aman and Alberti, Silas and Zinsley, Dylan and Zou, James and Rudra, Atri and Ré, Christopher},
journal={arXiv:2402.18668},
year={2024}
}
```
Please reach out to [email protected], [email protected], and [email protected] with questions. |
juexzz/INTACT-pi0-finetune-bridge | juexzz | 2025-06-15T19:18:26Z | 2 | 0 | null | [
"safetensors",
"robotics",
"arxiv:2410.24164",
"arxiv:2506.09930",
"base_model:lerobot/pi0",
"base_model:finetune:lerobot/pi0",
"license:apache-2.0",
"region:us"
] | robotics | 2025-06-15T02:21:01Z | ---
license: apache-2.0
base_model:
- lerobot/pi0
pipeline_tag: robotics
---
# INTACT Probing Suite: Pi0 Fine-tuned on BridgeV2
> 📦 **This model is part of the [INTACT Probing Suite Collection](https://huggingface.co/collections/ai4ce/intact-probing-suite-684e5601e9ed640fdd9b994b)**
> Explore other variants:
> - [Pi0 from scratch on BridgeV2](https://huggingface.co/juexzz/INTACT-pi0-scratch-bridge)
> - [Pi0 finetuned with paraphrase on BridgeV2](https://huggingface.co/juexzz/INTACT-pi0-finetune-rephrase-bridge)
## INTACT-pi-finetune-bridge
This repository contains a checkpoint of the Pi0 model ([HF implementation](https://huggingface.co/lerobot/pi0) | [Paper](https://arxiv.org/abs/2410.24164v1)) finetuned on the BridgeV2 dataset for robotic manipulation tasks.
The model is later used for testing on the [Simpler Environment](https://github.com/simpler-env/SimplerEnv) and our [INTACT](https://github.com/ai4ce/INT-ACT) Probing Suite for the generalization boundaries of VLA models.
**Paper**: [From Intention to Execution: Probing the Generalization Boundaries of Vision-Language-Action Models](https://arxiv.org/abs/2506.09930)
## Model Details
- **Base Model**: [lerobot/pi0](https://huggingface.co/lerobot/pi0)
- **Training Dataset**: [BridgeV2](https://rail-berkeley.github.io/bridgedata/)
- **Model Type**: Vision-Language-Action (VLA) model for robotics
- **Fine-tuning Method**: See our [paper](https://arxiv.org/abs/2506.09930)
- **Training Framework**: See our [repository](https://github.com/ai4ce/INT-ACT)
## Quick Start
### Usage in INTACT
```shell
git clone --recurse-submodules https://github.com/ai4ce/INT-ACT.git
cd INT-ACT
uv sync
source .venv/bin/activate
python
```
Or directly in python with Lerobot, see blow:
### Integration with LeRobot
First, install lerobot
```bash
pip install lerobot
```
Then
```python
import torch
from lerobot.common.policies.pi0.modeling_pi0 import Pi0Policy
# Load model
policy = Pi0Policy.from_pretrained("juexzz/INTACT-pi0-finetune-bridge")
# Inference
with torch.no_grad():
actions = policy.select_action(batch)
```
### Training Configuration
- **Training Steps**: 15 epochs ~22695 steps.
- **Batch Size**: 1024
- **Learning Rate**: 1e-5
- **Hardware**: 4 H100/A100
- **Input Modalities**: single image (to work with SimplerEnv), 1 language instruction, 1 robot state.
- **Output**: robot actions (delta EEF) with chunk size of 4.
For more details please refer to our [paper](https://arxiv.org/abs/2506.09930) and [code](https://github.com/ai4ce/INT-ACT)
## Evaluation
**Checkpoint choice**
After training 15 epochs, we sweep the checkpoint at epoch 1, 2, 3, 4, 5, 10, 15 for performance on the original 4 Bridge tasks in the SimplerEnv, and choose the checkpoint with *best average performance* for each of the three Pi0 variants.
Therefore, you may still get a better success rate for a specific task at other checkpoints.
As a result, the best checkpoint for this pi0 finetune model is at step 7565 (epoch 5).
The comparison of their performance on Simpler are shown below.
### Performance Comparison on SimplerEnv
**Success rate** comparison on the SimplerEnv with other pi0 variants and some other baselines experimented in our INTACT suite.
For a more detailed comparison, please refer to the [paper](https://arxiv.org/abs/2506.09930).
| Model | carrot_on_plate | eggplant_in_basket | stack_cube | spoon_on_towel |
|-------|-----------------|-------------------|------------|----------------|
| **Pi0 finetune (This Model)** | 0.361 | 0.819 | 0.264 | 0.458 |
| [Pi0 finetune rephrase](https://huggingface.co/juexzz/INTACT-pi0-finetune-rephrase-bridge) | 0.500 | 0.944 | 0.222 | 0.597 |
| [Pi0 scratch](https://huggingface.co/juexzz/INTACT-pi0-scratch-bridge) | 0.542 | 0.903 | 0.403 | 0.875 |
| Spatial VLA | 0.125 | 0.958 | 0.292 | 0.208 |
| Magma | 0.250 | 0.611 | 0.097 | 0.208 |
| Octo Small | 0.014 | 0.097 | 0.000 | 0.097 |
| Octo Base | 0.014 | 0.306 | 0.000 | 0.014 |
## Citation
If you use this model in your research, please cite:
```bibtex
@article{fang2025intention,
title={From Intention to Execution: Probing the Generalization Boundaries of Vision-Language-Action Models},
author={Fang, Irving and Zhang, Juexiao and Tong, Shengbang and Feng, Chen},
journal={arXiv preprint arXiv:2506.09930},
year={2025}
}
```
## Related Work
- **Pi0 (official)**: [pi0 (JAX)](https://github.com/Physical-Intelligence/openpi)
- **Base Model (Pi0 HF)**: [lerobot/pi0](https://huggingface.co/lerobot/pi0)
- **Dataset**: [BridgeV2](https://bridge-v2.github.io/)
- **Framework**: [LeRobot](https://github.com/huggingface/lerobot)
- **Simpler Environment**: [SimplerEnv](https://github.com/simpler-env/SimplerEnv)
- **Open-source Pi0 Implementation by Allen Ren**: [open-pi-zero](https://github.com/allenzren/open-pi-zero)
## License
This model is released under the Apache 2.0 license. Please see the base model's license for any additional restrictions.
## Support
For questions about this model:
- 📧 Open an issue in this repository
- 💬 Discussion tab for community questions
- 📖 Check our [paper](https://arxiv.org/abs/2506.09930) for technical details
---
*Last updated: June 2025* |
Newark-Airport-Indian-Student-Videos/FULL.VIDEO.Newark.Airport.Indian.Student.Viral.Video.Official | Newark-Airport-Indian-Student-Videos | 2025-06-15T19:16:55Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:16:33Z | <animated-image data-catalyst=""><a href="https://tinyurl.com/5ye5v3bc?dfhgKasbonStudiosdfg" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a>
|
Katrina-Lim-Viral-Kiffy-Viral-videos-usa/Original.Full.Clip.Katrina.Lim.Viral.Video.Official | Katrina-Lim-Viral-Kiffy-Viral-videos-usa | 2025-06-15T19:15:44Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:15:26Z |
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk">🌐 𝖢𝖫𝖨𝖢𝖪 𝖧𝖤𝖱𝖤 🟢==►► 𝖶𝖠𝖳𝖢𝖧 𝖭𝖮𝖶</a>
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk">🔴 CLICK HERE 🌐==►► Download Now)</a>
<a data-target="animated-image.originalLink" rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk"><img data-target="animated-image.originalImage" style="max-width: 100%; display: inline-block;" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif"></a> |
MinaMila/gemma_2b_unlearned_2nd_5e-7_1.0_0.15_0.15_0.75_epoch1 | MinaMila | 2025-06-15T19:15:38Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T19:13:39Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
TV-nulook-india-viral-videos-Original/18.video.Clip.nulook.india.Viral.Video.Leaks.Official | TV-nulook-india-viral-videos-Original | 2025-06-15T19:14:42Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:13:56Z | <animated-image data-catalyst=""><a href="https://sexleakedviral.com/new-leaked-video/?news-viral-video" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a> |
Katrina-Lim-Viral-Kiffy-Viral-videos-usa/FULL.VIDEO.Katrina.Lim.Video.Viral.Tutorial.Official | Katrina-Lim-Viral-Kiffy-Viral-videos-usa | 2025-06-15T19:14:29Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:14:08Z |
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk">🌐 𝖢𝖫𝖨𝖢𝖪 𝖧𝖤𝖱𝖤 🟢==►► 𝖶𝖠𝖳𝖢𝖧 𝖭𝖮𝖶</a>
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk">🔴 CLICK HERE 🌐==►► Download Now)</a>
<a data-target="animated-image.originalLink" rel="nofollow" href="https://viralvideoclipe.store/viral-videos/?kk"><img data-target="animated-image.originalImage" style="max-width: 100%; display: inline-block;" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif"></a>
|
07-katrina-lim-viral/8K.Link.katrina.lim.viral.video.Original.video.today | 07-katrina-lim-viral | 2025-06-15T19:13:44Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:06:45Z | <a data-target="animated-image.originalLink" rel="nofollow" href="https://viralvideoclipe.store/viral-videos/"><img data-target="animated-image.originalImage" style="max-width: 100%; display: inline-block;" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif"></a>
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/">🌐 𝖢𝖫𝖨𝖢𝖪 𝖧𝖤𝖱𝖤 🟢==►► 𝖶𝖠𝖳𝖢𝖧 𝖭𝖮𝖶</a>
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/">🔴 CLICK HERE 🌐==►► Download Now)</a>
|
18-bleep-Viral-Videos/ULL.VIDEO.bleep.Viral.Video.Tutorial.Official | 18-bleep-Viral-Videos | 2025-06-15T19:12:49Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:10:05Z | [<img alt="fsd" src="http://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/?bleep)
[🔴 ➤►𝐂𝐥𝐢𝐤 𝐇𝐞𝐫𝐞 𝐭𝐨👉👉 (𝐅𝐮𝐥𝐥 𝐯𝐢𝐝𝐞𝐨 𝐋𝐢𝐧𝐤 )](https://videohere.top/?bleep)
[🔴 ➤►𝐂𝐥𝐢𝐤 𝐇𝐞𝐫𝐞 𝐭𝐨👉👉 (𝐅𝐮𝐥𝐥 𝐯𝐢𝐝𝐞𝐨 𝐋𝐢𝐧𝐤 )](https://videohere.top/?bleep) |
Manal0809/MedQA_Mistral_7b_Instructive_KG | Manal0809 | 2025-06-15T19:12:39Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"mistral",
"trl",
"en",
"base_model:unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
"base_model:finetune:unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:12:29Z | ---
base_model: unsloth/mistral-7b-instruct-v0.3-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- mistral
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** Manal0809
- **License:** apache-2.0
- **Finetuned from model :** unsloth/mistral-7b-instruct-v0.3-bnb-4bit
This mistral model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
gradientrouting-spar/mc14_badmed_dpo_dsd-1_msd-1_atc-0.45_ldpo-6_seed_1 | gradientrouting-spar | 2025-06-15T19:11:26Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:11:12Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
saujasv/pixtral-coco-6-images-listener-2 | saujasv | 2025-06-15T19:09:53Z | 0 | 0 | peft | [
"peft",
"safetensors",
"arxiv:1910.09700",
"base_model:saujasv/pixtral-12b",
"base_model:adapter:saujasv/pixtral-12b",
"region:us"
] | null | 2025-06-14T16:47:04Z | ---
base_model: saujasv/pixtral-12b
library_name: peft
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
### Framework versions
- PEFT 0.13.1 |
MinaMila/gemma_2b_unlearned_2nd_5e-7_1.0_0.15_0.25_0.05_epoch2 | MinaMila | 2025-06-15T19:07:16Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T19:05:19Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
VIDEOS-18-parveen-viral-video-tv/New.tutorial.parveen.Viral.Video.Leaks.Official | VIDEOS-18-parveen-viral-video-tv | 2025-06-15T19:06:58Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:06:36Z | <animated-image data-catalyst=""><a href="https://tinyurl.com/5ye5v3bc?dfhgKasbonStudiosdfg" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a>
|
07-katrina-lim-viral/ORIGINAL.FULL.VIDEO.Katrina.Lim.Viral.Video.Leaks.Official-videos | 07-katrina-lim-viral | 2025-06-15T19:04:36Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:04:15Z | <a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/">🌐 𝖢𝖫𝖨𝖢𝖪 𝖧𝖤𝖱𝖤 🟢==►► 𝖶𝖠𝖳𝖢𝖧 𝖭𝖮𝖶</a>
<a rel="nofollow" href="https://viralvideoclipe.store/viral-videos/">🔴 CLICK HERE 🌐==►► Download Now)</a>
<a data-target="animated-image.originalLink" rel="nofollow" href="https://viralvideoclipe.store/viral-videos/"><img data-target="animated-image.originalImage" style="max-width: 100%; display: inline-block;" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif"></a>
|
gradientrouting-spar/horizontal_5_proxy_ntrain_25_ntrig_9_animals_3x3_seed_1_seed_25_seed_2_20250615_185444 | gradientrouting-spar | 2025-06-15T19:04:05Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T19:03:58Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
meezo-fun-18/Original.FULL.VIDEO.LINK.meezo-fun.meezo-fun.Video.Leaks.Official | meezo-fun-18 | 2025-06-15T19:03:57Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T19:00:43Z | [🔴 ➤►𝐂𝐥𝐢𝐤 𝐇𝐞𝐫𝐞 𝐭𝐨👉👉 (𝐅𝐮𝐥𝐥 𝐯𝐢𝐝𝐞𝐨 𝐋𝐢𝐧𝐤 )](https://videohere.top/?meezo-fun)
[►✅ 𝘾𝙇𝙄𝘾𝙆 𝙃𝙀𝙍𝙀 ==►► 𝙁𝙪𝙡𝙡 𝙑𝙞𝙙𝙚𝙤❤️❤️⬇️⬇️](https://videohere.top/?meezo-fun)
[<img alt="fsd" src="http://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/?meezo-fun) |
Kaidiyar/distilbert-base-uncased-finetuned-imdb | Kaidiyar | 2025-06-15T19:03:14Z | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"distilbert",
"fill-mask",
"generated_from_trainer",
"base_model:distilbert/distilbert-base-uncased",
"base_model:finetune:distilbert/distilbert-base-uncased",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2025-06-15T18:50:50Z | ---
library_name: transformers
license: apache-2.0
base_model: distilbert-base-uncased
tags:
- generated_from_trainer
model-index:
- name: distilbert-base-uncased-finetuned-imdb
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-base-uncased-finetuned-imdb
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 2.4894
- Model Preparation Time: 0.0017
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 3.0
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Model Preparation Time |
|:-------------:|:-----:|:----:|:---------------:|:----------------------:|
| 2.6838 | 1.0 | 157 | 2.5094 | 0.0017 |
| 2.5878 | 2.0 | 314 | 2.4502 | 0.0017 |
| 2.5279 | 3.0 | 471 | 2.4819 | 0.0017 |
### Framework versions
- Transformers 4.52.4
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.21.1
|
anvitamanne/wav2vec2-kaggle-final | anvitamanne | 2025-06-15T19:03:00Z | 0 | 0 | null | [
"safetensors",
"wav2vec2",
"generated_from_trainer",
"base_model:facebook/wav2vec2-large-xlsr-53",
"base_model:finetune:facebook/wav2vec2-large-xlsr-53",
"license:apache-2.0",
"region:us"
] | null | 2025-06-12T15:57:00Z | ---
license: apache-2.0
base_model: facebook/wav2vec2-large-xlsr-53
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: wav2vec2-kaggle-final
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# wav2vec2-kaggle-final
This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 513.6357
- Wer: 0.4037
- Cer: 0.1660
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 15
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|:-------------:|:-----:|:-----:|:---------------:|:------:|:------:|
| 1711.3334 | 0.86 | 1000 | 768.0388 | 0.8588 | 0.3197 |
| 672.7345 | 1.72 | 2000 | 522.8168 | 0.6091 | 0.2202 |
| 574.6395 | 2.58 | 3000 | 495.6673 | 0.5407 | 0.2048 |
| 518.2652 | 3.44 | 4000 | 472.2298 | 0.5068 | 0.1910 |
| 485.7279 | 4.3 | 5000 | 448.1584 | 0.4797 | 0.1835 |
| 456.6944 | 5.17 | 6000 | 459.5286 | 0.4703 | 0.1805 |
| 440.0209 | 6.03 | 7000 | 490.1409 | 0.4549 | 0.1780 |
| 424.1306 | 6.89 | 8000 | 458.7100 | 0.4455 | 0.1754 |
| 413.0438 | 7.75 | 9000 | 446.2839 | 0.4371 | 0.1735 |
| 382.4416 | 8.61 | 10000 | 499.2264 | 0.4338 | 0.1728 |
| 370.5859 | 9.47 | 11000 | 489.0332 | 0.4243 | 0.1692 |
| 352.6317 | 10.33 | 12000 | 491.2080 | 0.4133 | 0.1664 |
| 371.7963 | 11.19 | 13000 | 464.0348 | 0.4109 | 0.1657 |
| 343.1062 | 12.05 | 14000 | 488.7343 | 0.4134 | 0.1676 |
| 332.8081 | 12.91 | 15000 | 518.1512 | 0.4044 | 0.1649 |
| 323.9083 | 13.78 | 16000 | 507.0865 | 0.4072 | 0.1667 |
| 323.2671 | 14.64 | 17000 | 513.6357 | 0.4037 | 0.1660 |
### Framework versions
- Transformers 4.38.2
- Pytorch 2.1.0+cu118
- Datasets 3.6.0
- Tokenizers 0.15.2
|
gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in_qnli | gokulsrinivasagan | 2025-06-15T19:00:18Z | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"bert",
"text-classification",
"generated_from_trainer",
"en",
"dataset:glue",
"base_model:gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in",
"base_model:finetune:gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2025-06-15T18:48:33Z | ---
library_name: transformers
language:
- en
license: apache-2.0
base_model: gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in
tags:
- generated_from_trainer
datasets:
- glue
metrics:
- accuracy
model-index:
- name: tinybert_base_train_book_ent_15p_s_init_kd_a_in_qnli
results:
- task:
name: Text Classification
type: text-classification
dataset:
name: GLUE QNLI
type: glue
args: qnli
metrics:
- name: Accuracy
type: accuracy
value: 0.8390993959362988
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# tinybert_base_train_book_ent_15p_s_init_kd_a_in_qnli
This model is a fine-tuned version of [gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in](https://huggingface.co/gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in) on the GLUE QNLI dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3708
- Accuracy: 0.8391
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 256
- eval_batch_size: 256
- seed: 10
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 50
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| 0.5081 | 1.0 | 410 | 0.4320 | 0.8036 |
| 0.413 | 2.0 | 820 | 0.3758 | 0.8342 |
| 0.3639 | 3.0 | 1230 | 0.3708 | 0.8391 |
| 0.3163 | 4.0 | 1640 | 0.3786 | 0.8446 |
| 0.2707 | 5.0 | 2050 | 0.4099 | 0.8364 |
| 0.2332 | 6.0 | 2460 | 0.4355 | 0.8407 |
| 0.1964 | 7.0 | 2870 | 0.4571 | 0.8347 |
| 0.1673 | 8.0 | 3280 | 0.5541 | 0.8155 |
### Framework versions
- Transformers 4.51.2
- Pytorch 2.6.0+cu126
- Datasets 3.5.0
- Tokenizers 0.21.1
|
MinaMila/gemma_2b_unlearned_2nd_5e-7_1.0_0.15_0.25_0.05_epoch1 | MinaMila | 2025-06-15T18:59:16Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T18:57:26Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
swdq/stock-prices-model | swdq | 2025-06-15T18:57:18Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T18:55:58Z | claude@189b6b4e894a:/app$ python3 '/app/inference.py'
Using checkpoint: checkpoints/last.ckpt
Model loaded from checkpoints/last.ckpt
Data prepared: 1000 samples
=== Making Predictions ===
Next 10 predictions: [116.53201066 116.37892322 116.04956024 115.62472269 115.15303941
114.6622934 114.16789415 113.67827708 113.19807399 112.72984611]
=== Model Evaluation ===
MSE: 7.7239
MAE: 2.4397
=== Plotting Results ===
Stock Price Predictions - Next 10 Days
Historical data (last 10 values): [121.14573944 119.10400458 118.85682719 118.17484295 117.17322294
116.89212264 118.68980917 119.33065203 118.75947304 119.33205582]
Predictions: [116.53201066 116.37892322 116.04956024 115.62472269 115.15303941
114.6622934 114.16789415 113.67827708 113.19807399 112.72984611]
(Plot would be saved to predictions.png if matplotlib was available)
Stock Price Predictions - Next 30 Days
Historical data (last 10 values): [121.14573944 119.10400458 118.85682719 118.17484295 117.17322294
116.89212264 118.68980917 119.33065203 118.75947304 119.33205582]
Predictions: [118.31780125 117.89574121 117.38160134 116.82251489 116.24679503
115.67077976 115.10363774 114.55039909 114.01377537 113.49513293
112.99504679 112.51360925 112.05058821 111.60554316 111.17787799
110.76690689 110.37187547 109.99200297 109.62649281 109.27455894
108.93542852 108.60836035 108.29264488 107.98761477 107.69263435
107.40711275 107.13050398 106.8622884 106.60199122 106.34917196]
(Plot would be saved to long_predictions.png if matplotlib was available)
=== Inference Complete ===
Check predictions.png and long_predictions.png for visualizations
claude@189b6b4e894a:/app$ |
Shrav20/sparql-mistral-lora-4bit | Shrav20 | 2025-06-15T18:56:04Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T18:56:00Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
duchao1210/DPO_Qwen25_3B_64_0_5000kmaplr1e-7 | duchao1210 | 2025-06-15T18:53:22Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"text-generation-inference",
"unsloth",
"conversational",
"en",
"base_model:duchao1210/qwen_2.5_3B_5k_r128",
"base_model:finetune:duchao1210/qwen_2.5_3B_5k_r128",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T18:51:55Z | ---
base_model: duchao1210/qwen_2.5_3B_5k_r128
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** duchao1210
- **License:** apache-2.0
- **Finetuned from model :** duchao1210/qwen_2.5_3B_5k_r128
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in_mrpc | gokulsrinivasagan | 2025-06-15T18:48:06Z | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"bert",
"text-classification",
"generated_from_trainer",
"en",
"dataset:glue",
"base_model:gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in",
"base_model:finetune:gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2025-06-15T18:47:19Z | ---
library_name: transformers
language:
- en
license: apache-2.0
base_model: gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in
tags:
- generated_from_trainer
datasets:
- glue
metrics:
- accuracy
- f1
model-index:
- name: tinybert_base_train_book_ent_15p_s_init_kd_a_in_mrpc
results:
- task:
name: Text Classification
type: text-classification
dataset:
name: GLUE MRPC
type: glue
args: mrpc
metrics:
- name: Accuracy
type: accuracy
value: 0.7254901960784313
- name: F1
type: f1
value: 0.819935691318328
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# tinybert_base_train_book_ent_15p_s_init_kd_a_in_mrpc
This model is a fine-tuned version of [gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in](https://huggingface.co/gokulsrinivasagan/tinybert_base_train_book_ent_15p_s_init_kd_a_in) on the GLUE MRPC dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5727
- Accuracy: 0.7255
- F1: 0.8199
- Combined Score: 0.7727
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 256
- eval_batch_size: 256
- seed: 10
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 50
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | Combined Score |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:--------------:|
| 0.6214 | 1.0 | 15 | 0.5932 | 0.7010 | 0.8100 | 0.7555 |
| 0.5853 | 2.0 | 30 | 0.5762 | 0.7157 | 0.8204 | 0.7681 |
| 0.5515 | 3.0 | 45 | 0.5727 | 0.7255 | 0.8199 | 0.7727 |
| 0.5201 | 4.0 | 60 | 0.5971 | 0.6985 | 0.7776 | 0.7381 |
| 0.4485 | 5.0 | 75 | 0.6437 | 0.6667 | 0.7527 | 0.7097 |
| 0.3653 | 6.0 | 90 | 0.6802 | 0.7010 | 0.7967 | 0.7488 |
| 0.2927 | 7.0 | 105 | 0.7607 | 0.6814 | 0.7610 | 0.7212 |
| 0.232 | 8.0 | 120 | 0.8599 | 0.7108 | 0.8007 | 0.7557 |
### Framework versions
- Transformers 4.51.2
- Pytorch 2.6.0+cu126
- Datasets 3.5.0
- Tokenizers 0.21.1
|
Varinder2110/f45662fb-3420-4734-bdbc-633f329f71ca | Varinder2110 | 2025-06-15T18:44:57Z | 0 | 0 | diffusers | [
"diffusers",
"flux",
"lora",
"replicate",
"text-to-image",
"en",
"base_model:black-forest-labs/FLUX.1-dev",
"base_model:adapter:black-forest-labs/FLUX.1-dev",
"license:other",
"region:us"
] | text-to-image | 2025-06-15T18:00:32Z | ---
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
language:
- en
tags:
- flux
- diffusers
- lora
- replicate
base_model: "black-forest-labs/FLUX.1-dev"
pipeline_tag: text-to-image
# widget:
# - text: >-
# prompt
# output:
# url: https://...
instance_prompt: TOK
---
# F45662Fb 3420 4734 Bdbc 633F329F71Ca
<Gallery />
## About this LoRA
This is a [LoRA](https://replicate.com/docs/guides/working-with-loras) for the FLUX.1-dev text-to-image model. It can be used with diffusers or ComfyUI.
It was trained on [Replicate](https://replicate.com/) using AI toolkit: https://replicate.com/ostris/flux-dev-lora-trainer/train
## Trigger words
You should use `TOK` to trigger the image generation.
## Run this LoRA with an API using Replicate
```py
import replicate
input = {
"prompt": "TOK",
"lora_weights": "https://huggingface.co/Varinder2110/f45662fb-3420-4734-bdbc-633f329f71ca/resolve/main/lora.safetensors"
}
output = replicate.run(
"black-forest-labs/flux-dev-lora",
input=input
)
for index, item in enumerate(output):
with open(f"output_{index}.webp", "wb") as file:
file.write(item.read())
```
## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```py
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('Varinder2110/f45662fb-3420-4734-bdbc-633f329f71ca', weight_name='lora.safetensors')
image = pipeline('TOK').images[0]
```
For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
## Training details
- Steps: 4000
- Learning rate: 0.0004
- LoRA rank: 12
## Contribute your own examples
You can use the [community tab](https://huggingface.co/Varinder2110/f45662fb-3420-4734-bdbc-633f329f71ca/discussions) to add images that show off what you’ve made with this LoRA.
|
MinaMila/gemma_2b_unlearned_2nd_5e-7_1.0_0.15_0.25_0.15_epoch1 | MinaMila | 2025-06-15T18:43:14Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T18:41:22Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
JosephTreitel/reddit-lora-V3 | JosephTreitel | 2025-06-15T18:39:48Z | 0 | 1 | null | [
"safetensors",
"mistral",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T18:18:18Z | ---
license: apache-2.0
---
|
trancuong253/grpo_final | trancuong253 | 2025-06-15T18:38:11Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"qwen2",
"trl",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T18:38:02Z | ---
base_model: unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** trancuong253
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
BootesVoid/cmbxwm6wh027lrdqs6c7udorq_cmbxwwd5a028mrdqsf4hpeuhh | BootesVoid | 2025-06-15T18:37:52Z | 0 | 0 | diffusers | [
"diffusers",
"flux",
"lora",
"replicate",
"text-to-image",
"en",
"base_model:black-forest-labs/FLUX.1-dev",
"base_model:adapter:black-forest-labs/FLUX.1-dev",
"license:other",
"region:us"
] | text-to-image | 2025-06-15T18:37:50Z | ---
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
language:
- en
tags:
- flux
- diffusers
- lora
- replicate
base_model: "black-forest-labs/FLUX.1-dev"
pipeline_tag: text-to-image
# widget:
# - text: >-
# prompt
# output:
# url: https://...
instance_prompt: JUSTTURNED18
---
# Cmbxwm6Wh027Lrdqs6C7Udorq_Cmbxwwd5A028Mrdqsf4Hpeuhh
<Gallery />
## About this LoRA
This is a [LoRA](https://replicate.com/docs/guides/working-with-loras) for the FLUX.1-dev text-to-image model. It can be used with diffusers or ComfyUI.
It was trained on [Replicate](https://replicate.com/) using AI toolkit: https://replicate.com/ostris/flux-dev-lora-trainer/train
## Trigger words
You should use `JUSTTURNED18` to trigger the image generation.
## Run this LoRA with an API using Replicate
```py
import replicate
input = {
"prompt": "JUSTTURNED18",
"lora_weights": "https://huggingface.co/BootesVoid/cmbxwm6wh027lrdqs6c7udorq_cmbxwwd5a028mrdqsf4hpeuhh/resolve/main/lora.safetensors"
}
output = replicate.run(
"black-forest-labs/flux-dev-lora",
input=input
)
for index, item in enumerate(output):
with open(f"output_{index}.webp", "wb") as file:
file.write(item.read())
```
## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```py
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('BootesVoid/cmbxwm6wh027lrdqs6c7udorq_cmbxwwd5a028mrdqsf4hpeuhh', weight_name='lora.safetensors')
image = pipeline('JUSTTURNED18').images[0]
```
For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
## Training details
- Steps: 2000
- Learning rate: 0.0004
- LoRA rank: 16
## Contribute your own examples
You can use the [community tab](https://huggingface.co/BootesVoid/cmbxwm6wh027lrdqs6c7udorq_cmbxwwd5a028mrdqsf4hpeuhh/discussions) to add images that show off what you’ve made with this LoRA.
|
mergekit-community/mergekit-slerp-tuqnqct | mergekit-community | 2025-06-15T18:37:10Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen3",
"text-generation",
"mergekit",
"merge",
"conversational",
"base_model:Qwen/Qwen3-8B",
"base_model:merge:Qwen/Qwen3-8B",
"base_model:deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
"base_model:merge:deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2025-06-15T18:31:05Z | ---
base_model:
- Qwen/Qwen3-8B
- deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
library_name: transformers
tags:
- mergekit
- merge
---
# merge
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
## Merge Details
### Merge Method
This model was merged using the [SLERP](https://en.wikipedia.org/wiki/Slerp) merge method.
### Models Merged
The following models were included in the merge:
* [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B)
* [deepseek-ai/DeepSeek-R1-0528-Qwen3-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B)
### Configuration
The following YAML configuration was used to produce this model:
```yaml
slices:
- sources:
- model: Qwen/Qwen3-8B
layer_range: [0, 35]
- model: deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
layer_range: [0, 35]
merge_method: slerp
base_model: Qwen/Qwen3-8B
parameters:
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 1]
- filter: mlp
value: [1, 0.5, 0.7, 0.3, 0]
- value: 0.5
dtype: bfloat16
```
|
vex1522/baseline-summary-model | vex1522 | 2025-06-15T18:35:59Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"bart",
"text2text-generation",
"arxiv:1910.09700",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2025-06-15T18:04:38Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
gradientrouting-spar/horizontal_2_proxy_ntrain_25_ntrig_9_negative_3x3_seed_1_seed_25_seed_2_seed_42_20250615_182610 | gradientrouting-spar | 2025-06-15T18:35:29Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T18:35:21Z | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
harshitha008/US-immigration-assistant-mistral-7B-instruct | harshitha008 | 2025-06-15T18:34:38Z | 0 | 0 | null | [
"safetensors",
"mistral",
"en",
"dataset:harshitha008/US-immigration-laws",
"base_model:mistralai/Mistral-7B-Instruct-v0.3",
"base_model:finetune:mistralai/Mistral-7B-Instruct-v0.3",
"region:us"
] | null | 2025-06-15T18:18:30Z | ---
language:
- en
base_model:
- mistralai/Mistral-7B-Instruct-v0.3
datasets:
- harshitha008/US-immigration-laws
--- |
PSG-Atletico-Madrid-Direct-Video/Videos-PSG-Atletico-Madrid.En.Direct.Streaming.Gratuit.Official | PSG-Atletico-Madrid-Direct-Video | 2025-06-15T18:31:47Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T18:31:15Z | <animated-image data-catalyst=""><a href="https://tinyurl.com/mrmpsap6?dfhgKasbonStudiosdfg" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a>
|
NPC-GEMMA-1/gemma-3-4b-text-to-sql | NPC-GEMMA-1 | 2025-06-15T18:31:17Z | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"base_model:google/gemma-3-4b-it",
"base_model:finetune:google/gemma-3-4b-it",
"endpoints_compatible",
"region:us"
] | null | 2025-06-15T17:30:10Z | ---
base_model: google/gemma-3-4b-it
library_name: transformers
model_name: gemma-3-4b-text-to-sql
tags:
- generated_from_trainer
- trl
- sft
licence: license
---
# Model Card for gemma-3-4b-text-to-sql
This model is a fine-tuned version of [google/gemma-3-4b-it](https://huggingface.co/google/gemma-3-4b-it).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="NPC-GEMMA-1/gemma-3-4b-text-to-sql", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.18.2
- Transformers: 4.53.0.dev0
- Pytorch: 2.7.1
- Datasets: 3.6.0
- Tokenizers: 0.21.1
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
GetnetWA/Bert_Question_Ans | GetnetWA | 2025-06-15T18:29:07Z | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"bert",
"question-answering",
"generated_from_trainer",
"base_model:google-bert/bert-base-uncased",
"base_model:finetune:google-bert/bert-base-uncased",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | question-answering | 2025-06-13T18:31:34Z | ---
library_name: transformers
license: apache-2.0
base_model: bert-base-uncased
tags:
- generated_from_trainer
model-index:
- name: Bert_Question_Ans
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# Bert_Question_Ans
This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on an unknown dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 1
### Training results
### Framework versions
- Transformers 4.52.4
- Pytorch 2.7.0+cu118
- Datasets 3.6.0
- Tokenizers 0.21.1
|
tirdodbehbehani/yahoo-bert-05_balanced_2_mask_aug | tirdodbehbehani | 2025-06-15T18:27:09Z | 0 | 0 | transformers | [
"transformers",
"safetensors",
"bert",
"text-classification",
"generated_from_trainer",
"base_model:tirdodbehbehani/yahoo-bert-32shot_stratified_augm_2",
"base_model:finetune:tirdodbehbehani/yahoo-bert-32shot_stratified_augm_2",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2025-06-15T17:22:19Z | ---
library_name: transformers
base_model: tirdodbehbehani/yahoo-bert-32shot_stratified_augm_2
tags:
- generated_from_trainer
metrics:
- accuracy
- precision
- recall
model-index:
- name: yahoo-bert-05_balanced_2_mask_aug
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# yahoo-bert-05_balanced_2_mask_aug
This model is a fine-tuned version of [tirdodbehbehani/yahoo-bert-32shot_stratified_augm_2](https://huggingface.co/tirdodbehbehani/yahoo-bert-32shot_stratified_augm_2) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 1.0359
- Accuracy: 0.6692
- Precision: 0.6737
- Recall: 0.6657
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 20
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:---------:|:------:|
| 1.226 | 1.0 | 492 | 1.0359 | 0.6692 | 0.6737 | 0.6657 |
| 0.7918 | 2.0 | 984 | 1.0909 | 0.6655 | 0.6654 | 0.6595 |
| 0.5589 | 3.0 | 1476 | 1.2076 | 0.6603 | 0.6555 | 0.6557 |
| 0.3879 | 4.0 | 1968 | 1.3394 | 0.6490 | 0.6531 | 0.6451 |
### Framework versions
- Transformers 4.52.4
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.21.1
|
VIDEOS-18-nulook-india-Hot-video/TV.nulook.india.viral.videos.Full.Clip.nulook.india.Viral.Video.Leaks.Official | VIDEOS-18-nulook-india-Hot-video | 2025-06-15T18:27:08Z | 0 | 0 | null | [
"region:us"
] | null | 2025-06-15T18:19:53Z | <animated-image data-catalyst=""><a href="https://sexleakedviral.com/new-leaked-video/?news-viral-video" rel="nofollow" data-target="animated-image.originalLink"><img src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" alt="Foo" data-canonical-src="https://static.wixstatic.com/media/b249f9_adac8f70fb3f45b88691696c77de18f3~mv2.gif" style="max-width: 100%; display: inline-block;" data-target="animated-image.originalImage"></a> |
Pouyatr/blob_Qwen2.5-0.5B-Instruct_BOSS_5001 | Pouyatr | 2025-06-15T18:25:56Z | 0 | 0 | peft | [
"peft",
"safetensors",
"arxiv:1910.09700",
"base_model:Qwen/Qwen2.5-0.5B-Instruct",
"base_model:adapter:Qwen/Qwen2.5-0.5B-Instruct",
"region:us"
] | null | 2025-06-15T15:18:58Z | ---
base_model: Qwen/Qwen2.5-0.5B-Instruct
library_name: peft
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
### Framework versions
- PEFT 0.14.0 |
afonsovalente9184/lum | afonsovalente9184 | 2025-06-15T18:25:34Z | 0 | 0 | null | [
"license:artistic-2.0",
"region:us"
] | null | 2025-06-15T18:25:29Z | ---
license: artistic-2.0
---
|
lucamoura5296/sv | lucamoura5296 | 2025-06-15T18:25:34Z | 0 | 0 | null | [
"license:artistic-2.0",
"region:us"
] | null | 2025-06-15T18:25:28Z | ---
license: artistic-2.0
---
|
Subsets and Splits