Update README.md
Browse files
README.md
CHANGED
@@ -1,109 +1,39 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
-
|
|
|
6 |
tags:
|
7 |
- llama-3.1
|
8 |
- instruction-tuned
|
9 |
-
- fine-tuned
|
10 |
-
- merged-lora
|
11 |
-
license: llama3.1
|
12 |
datasets:
|
13 |
- OpenAssistant/oasst1
|
14 |
- databricks/databricks-dolly-15k
|
15 |
- Open-Orca/OpenOrca
|
16 |
-
- mlabonne/open-perfectblend
|
17 |
-
- tatsu-lab/alpaca
|
18 |
-
model-index:
|
19 |
-
- name: utkmst/chimera-beta-test2-lora-merged
|
20 |
-
results:
|
21 |
-
- task:
|
22 |
-
type: text-generation
|
23 |
-
name: Text Generation
|
24 |
-
metrics:
|
25 |
-
- name: Training Loss
|
26 |
-
type: loss
|
27 |
-
value: 2.143046485595703
|
28 |
-
pipeline_tag: text-generation
|
29 |
---
|
30 |
|
|
|
31 |
# utkmst/chimera-beta-test2-lora-merged
|
32 |
|
33 |
## Model Description
|
34 |
-
This model
|
35 |
-
|
36 |
-
## Architecture
|
37 |
-
- **Base Model**: meta-llama/Llama-3.1-8B-Instruct
|
38 |
-
- **Size**: 8.03B parameters
|
39 |
-
- **Type**: Decoder-only transformer
|
40 |
-
- **Format**: SafeTensors (full precision)
|
41 |
|
42 |
## Training Details
|
|
|
43 |
- **Training Method**: LoRA fine-tuning followed by adapter merging
|
44 |
-
- **
|
45 |
-
- Rank: 8
|
46 |
-
- Alpha: 16
|
47 |
-
- Trainable modules: Attention layers and feed-forward networks
|
48 |
-
- **Training Hyperparameters**:
|
49 |
-
- Learning rate: 2e-4
|
50 |
-
- Batch size: 2
|
51 |
-
- Training epochs: 1
|
52 |
-
- Optimizer: AdamW with constant scheduler
|
53 |
|
54 |
-
|
55 |
-
The model was trained on a curated mixture of high-quality instruction datasets:
|
56 |
-
- OpenAssistant/oasst1: Human-generated conversations with AI assistants
|
57 |
-
- databricks/databricks-dolly-15k: Instruction-following examples
|
58 |
-
- Open-Orca/OpenOrca: Augmented training data based on GPT-4 generations
|
59 |
-
- mlabonne/open-perfectblend: A carefully balanced blend of open-source instruction data
|
60 |
-
- tatsu-lab/alpaca: Self-instructed data based on demonstrations
|
61 |
-
|
62 |
-
## Intended Use
|
63 |
-
This model is designed for:
|
64 |
-
- General purpose assistant capabilities
|
65 |
-
- Question answering and knowledge retrieval
|
66 |
-
- Creative content generation
|
67 |
-
- Instructional guidance
|
68 |
-
|
69 |
-
## Limitations
|
70 |
-
- Base model limitations including potential hallucinations and factual inaccuracies
|
71 |
-
- Limited context window compared to larger models
|
72 |
-
- Knowledge cutoff from the base Llama-3.1 model
|
73 |
-
- May exhibit biases present in training data
|
74 |
-
- Performance on specialized tasks may vary
|
75 |
|
76 |
## Usage with Transformers
|
77 |
|
78 |
```python
|
79 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
80 |
|
81 |
-
# Load model
|
82 |
model = AutoModelForCausalLM.from_pretrained("utkmst/chimera-beta-test2-lora-merged")
|
83 |
tokenizer = AutoTokenizer.from_pretrained("utkmst/chimera-beta-test2-lora-merged")
|
84 |
-
|
85 |
-
# Format prompt according to Llama 3.1 chat template
|
86 |
-
messages = [
|
87 |
-
{"role": "user", "content": "Tell me about the solar system."}
|
88 |
-
]
|
89 |
-
prompt = tokenizer.apply_chat_template(messages, tokenize=False)
|
90 |
-
|
91 |
-
# Generate response
|
92 |
-
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
93 |
-
outputs = model.generate(
|
94 |
-
inputs["input_ids"],
|
95 |
-
max_new_tokens=512,
|
96 |
-
temperature=0.7,
|
97 |
-
top_p=0.9,
|
98 |
-
)
|
99 |
-
response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
|
100 |
-
print(response)
|
101 |
```
|
102 |
-
|
103 |
-
## Quantized Version
|
104 |
-
|
105 |
-
A quantized GGUF version of this model is also available at utkmst/chimera-beta-test2-lora-merged-Q4_K_M-GGUF for deployment in resource-constrained environments.
|
106 |
-
|
107 |
## License
|
108 |
|
109 |
-
This model inherits the license from Meta's Llama 3.1.
|
|
|
1 |
---
|
2 |
+
language: en
|
3 |
+
library_name: transformers
|
4 |
+
license: llama3.1
|
5 |
+
base_model: meta-llama/Llama-3.1-8B-Instruct
|
6 |
+
pipeline_tag: text-generation
|
7 |
tags:
|
8 |
- llama-3.1
|
9 |
- instruction-tuned
|
|
|
|
|
|
|
10 |
datasets:
|
11 |
- OpenAssistant/oasst1
|
12 |
- databricks/databricks-dolly-15k
|
13 |
- Open-Orca/OpenOrca
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
---
|
15 |
|
16 |
+
|
17 |
# utkmst/chimera-beta-test2-lora-merged
|
18 |
|
19 |
## Model Description
|
20 |
+
This model is a fine-tuned version of Meta's Llama-3.1-8B-Instruct model, created through LoRA fine-tuning on multiple instruction datasets, followed by merging the adapter weights with the base model.
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
## Training Details
|
23 |
+
- **Base Model**: meta-llama/Llama-3.1-8B-Instruct
|
24 |
- **Training Method**: LoRA fine-tuning followed by adapter merging
|
25 |
+
- **Datasets Used**: OpenAssistant/oasst1, databricks/databricks-dolly-15k, Open-Orca/OpenOrca, and others
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
+
(You can add more details from your original card here if desired)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
## Usage with Transformers
|
30 |
|
31 |
```python
|
32 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
33 |
|
|
|
34 |
model = AutoModelForCausalLM.from_pretrained("utkmst/chimera-beta-test2-lora-merged")
|
35 |
tokenizer = AutoTokenizer.from_pretrained("utkmst/chimera-beta-test2-lora-merged")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
```
|
|
|
|
|
|
|
|
|
|
|
37 |
## License
|
38 |
|
39 |
+
This model inherits the license from Meta's Llama 3.1.
|