Create README.md

README.md

@@ -0,0 +1,460 @@
+---
+tags:
+- vllm
+- vision
+- w4a16
+license: apache-2.0
+license_link: >-
+  https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md
+language:
+- en
+base_model: Qwen/Qwen2.5-VL-3B-Instruct
+library_name: transformers
+---
+
+# Qwen2.5-VL-3B-Instruct-quantized-w4a16
+
+## Model Overview
+- **Model Architecture:** Qwen/Qwen2.5-VL-3B-Instruct
+  - **Input:** Vision-Text
+  - **Output:** Text
+- **Model Optimizations:**
+  - **Weight quantization:** INT4
+  - **Activation quantization:** FP16
+- **Release Date:** 2/24/2025
+- **Version:** 1.0
+- **Model Developers:** Neural Magic
+
+Quantized version of [Qwen/Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights of [Qwen/Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) to INT8 data type, ready for inference with vLLM >= 0.5.2.
+
+## Deployment
+
+### Use with vLLM
+
+This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below.
+
+```python
+from vllm.assets.image import ImageAsset
+from vllm import LLM, SamplingParams
+
+# prepare model
+llm = LLM(
+    model="neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16",
+    trust_remote_code=True,
+    max_model_len=4096,
+    max_num_seqs=2,
+)
+
+# prepare inputs
+question = "What is the content of this image?"
+inputs = {
+    "prompt": f"<|user|>\n<|image_1|>\n{question}<|end|>\n<|assistant|>\n",
+    "multi_modal_data": {
+        "image": ImageAsset("cherry_blossom").pil_image.convert("RGB")
+    },
+}
+
+# generate response
+print("========== SAMPLE GENERATION ==============")
+outputs = llm.generate(inputs, SamplingParams(temperature=0.2, max_tokens=64))
+print(f"PROMPT  : {outputs[0].prompt}")
+print(f"RESPONSE: {outputs[0].outputs[0].text}")
+print("==========================================")
+```
+
+vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
+
+## Creation
+
+This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below as part a multimodal announcement blog.
+
+<details>
+  <summary>Model Creation Code</summary>
+  
+```python
+import base64
+from io import BytesIO
+import torch
+from datasets import load_dataset
+from qwen_vl_utils import process_vision_info
+from transformers import AutoProcessor
+from llmcompressor.modifiers.quantization import GPTQModifier
+from llmcompressor.transformers import oneshot
+from llmcompressor.transformers.tracing import (
+    TraceableQwen2_5_VLForConditionalGeneration,
+)
+from compressed_tensors.quantization import QuantizationArgs, QuantizationType, QuantizationStrategy, ActivationOrdering, QuantizationScheme
+
+# Load model.
+model_id = "Qwen/Qwen2.5-VL-3B-Instruct"
+
+model = TraceableQwen2_5_VLForConditionalGeneration.from_pretrained(
+    model_id,
+    device_map="auto",
+    torch_dtype="auto",
+)
+processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
+
+# Oneshot arguments
+DATASET_ID = "lmms-lab/flickr30k"
+DATASET_SPLIT = {"calibration": "test[:512]"}
+NUM_CALIBRATION_SAMPLES = 512
+MAX_SEQUENCE_LENGTH = 2048
+
+# Load dataset and preprocess.
+ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
+ds = ds.shuffle(seed=42)
+dampening_frac=0.01
+
+# Apply chat template and tokenize inputs.
+def preprocess_and_tokenize(example):
+    # preprocess
+    buffered = BytesIO()
+    example["image"].save(buffered, format="PNG")
+    encoded_image = base64.b64encode(buffered.getvalue())
+    encoded_image_text = encoded_image.decode("utf-8")
+    base64_qwen = f"data:image;base64,{encoded_image_text}"
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "image", "image": base64_qwen},
+                {"type": "text", "text": "What does the image show?"},
+            ],
+        }
+    ]
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+
+    # tokenize
+    return processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=False,
+        max_length=MAX_SEQUENCE_LENGTH,
+        truncation=True,
+    )
+ds = ds.map(preprocess_and_tokenize, remove_columns=ds["calibration"].column_names)
+
+# Define a oneshot data collator for multimodal inputs.
+def data_collator(batch):
+    assert len(batch) == 1
+    return {key: torch.tensor(value) for key, value in batch[0].items()}
+
+recipe = GPTQModifier(
+    targets="Linear",
+    config_groups={
+        "config_group": QuantizationScheme(
+            targets=["Linear"],
+            weights=QuantizationArgs(
+                num_bits=4,
+                type=QuantizationType.INT,
+                strategy=QuantizationStrategy.GROUP,
+                group_size=128,
+                symmetric=True,
+                dynamic=False,
+                actorder=ActivationOrdering.WEIGHT,
+            ),
+        ),
+    },
+    sequential_targets=["Qwen2_5_VLDecoderLayer"],
+    ignore=["lm_head", "re:visual.*"],
+    update_size=NUM_CALIBRATION_SAMPLES,
+    dampening_frac=dampening_frac
+)
+
+SAVE_DIR=f"{model_id.split('/')[1]}-quantized.w4a16"
+
+# Perform oneshot
+oneshot(
+    model=model,
+    tokenizer=model_id,
+    dataset=ds,
+    recipe=recipe,
+    max_seq_length=MAX_SEQUENCE_LENGTH,
+    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
+    trust_remote_code_model=True,
+    data_collator=data_collator,
+    output_dir=SAVE_DIR
+)
+
+```
+</details>
+
+## Evaluation
+
+The model was evaluated on OpenLLM Leaderboard [V1](https://huggingface.co/spaces/open-llm-leaderboard-old/open_llm_leaderboard), OpenLLM Leaderboard [V2](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/) and on [HumanEval](https://github.com/neuralmagic/evalplus), using the following commands:
+
+<details>
+<summary>Evaluation Commands</summary>
+
+```
+```
+
+</details>
+
+### Accuracy
+
+## Inference Performance
+
+
+This model achieves up to xxx speedup in single-stream deployment and up to xxx speedup in multi-stream asynchronous deployment, depending on hardware and use-case scenario.
+The following performance benchmarks were conducted with [vLLM](https://docs.vllm.ai/en/latest/) version 0.7.2, and [GuideLLM](https://github.com/neuralmagic/guidellm).
+
+<details>
+<summary>Benchmarking Command</summary>
+```
+  guidellm --model neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16 --target "http://localhost:8000/v1" --data-type emulated --data prompt_tokens=<prompt_tokens>,generated_tokens=<generated_tokens>,images=<num_images>,width=<image_width>,height=<image_height> --max seconds 120 --backend aiohttp_server
+```
+
+</details>
+
+### Single-stream performance (measured with vLLM version 0.7.2)
+
+<table border="1" class="dataframe">
+  <thead>
+    <tr>
+      <th></th>
+      <th></th>
+      <th></th>
+      <th style="text-align: center;" colspan="2" >Document Visual Question Answering<br>1680W x 2240H<br>64/128</th>
+      <th style="text-align: center;" colspan="2" >Visual Reasoning <br>640W x 480H<br>128/128</th>
+      <th style="text-align: center;" colspan="2" >Image Captioning<br>480W x 360H<br>0/128</th>
+    </tr>
+    <tr>
+      <th>Hardware</th>
+      <th>Model</th>
+      <th>Average Cost Reduction</th>
+      <th>Latency (s)</th>
+      <th>QPD</th>
+      <th>Latency (s)th>
+      <th>QPD</th>
+      <th>Latency (s)</th>
+      <th>QPD</th>
+    </tr>
+  </thead>
+  <tbody style="text-align: center">
+    <tr>
+      <th rowspan="3" valign="top">A6000x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td></td>
+      <td>3.1</td>
+      <td>1454</td>
+      <td>1.8</td>
+      <td>2546</td>
+      <td>1.7</td>
+      <td>2610</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w8a8</th>
+      <td>1.27</td>
+      <td>2.6</td>
+      <td>1708</td>
+      <td>1.3</td>
+      <td>3340</td>
+      <td>1.3</td>
+      <td>3459</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.57</td>
+      <td>2.4</td>
+      <td>1886</td>
+      <td>1.0</td>
+      <td>4409</td>
+      <td>1.0</td>
+      <td>4409</td>
+    </tr>
+    <tr>
+      <th rowspan="3" valign="top">A100x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td></td>
+      <td>2.2</td>
+      <td>920</td>
+      <td>1.3</td>
+      <td>1603</td>
+      <td>1.2</td>
+      <td>1636</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w8a8</th>
+      <td>1.09</td>
+      <td>2.1</td>
+      <td>975</td>
+      <td>1.2</td>
+      <td>1743</td>
+      <td>1.1</td>
+      <td>1814</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.20</td>
+      <td>2.0</td>
+      <td>1011</td>
+      <td>1.0</td>
+      <td>2015</td>
+      <td>1.0</td>
+      <td>2012</td>
+    </tr>
+    <tr>
+      <th rowspan="3" valign="top">H100x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td>1.5</td>
+      <td>740</td>
+      <td>0.9</td>
+      <td>1221</td>
+      <td>0.9</td>
+      <td>1276</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-FP8-Dynamic</th>
+      <td>1.06</td>
+      <td>1.4</td>
+      <td>768</td>
+      <td>0.9</td>
+      <td>1276</td>
+      <td>0.8</td>
+      <td>1399</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.24</td>
+      <td>0.9</td>
+      <td>1219</td>
+      <td>0.9</td>
+      <td>1270</td>
+      <td>0.8</td>
+      <td>1304</td>
+    </tr>
+  </tbody>
+</table>
+ 
+
+
+### Multi-stream asynchronous performance (measured with vLLM version 0.7.2)
+
+<table border="1" class="dataframe">
+  <thead>
+    <tr>
+      <th></th>
+      <th></th>
+      <th></th>
+      <th style="text-align: center;" colspan="2" >Document Visual Question Answering<br>1680W x 2240H<br>64/128</th>
+      <th style="text-align: center;" colspan="2" >Visual Reasoning <br>640W x 480H<br>128/128</th>
+      <th style="text-align: center;" colspan="2" >Image Captioning<br>480W x 360H<br>0/128</th>
+    </tr>
+    <tr>
+      <th>Hardware</th>
+      <th>Model</th>
+      <th>Average Cost Reduction</th>
+      <th>Maximum throughput (QPS)</th>
+      <th>QPD</th>
+      <th>Maximum throughput (QPS)</th>
+      <th>QPD</th>
+      <th>Maximum throughput (QPS)</th>
+      <th>QPD</th>
+    </tr>
+  </thead>
+  <tbody style="text-align: center">
+    <tr>
+      <th rowspan="3" valign="top">A6000x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td></td>
+      <td>0.5</td>
+      <td>2405</td>
+      <td>2.6</td>
+      <td>11889</td>
+      <td>2.9</td>
+      <td>12909</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w8a8</th>
+      <td>1.26</td>
+      <td>0.6</td>
+      <td>2725</td>
+      <td>3.4</td>
+      <td>15162</td>
+      <td>3.9</td>
+      <td>17673</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.39</td>
+      <td>0.6</td>
+      <td>2548</td>
+      <td>3.9</td>
+      <td>17437</td>
+      <td>4.7</td>
+      <td>21223</td>
+    </tr>
+    <tr>
+      <th rowspan="3" valign="top">A100x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td></td>
+      <td>0.8</td>
+      <td>1663</td>
+      <td>3.9</td>
+      <td>7899</td>
+      <td>4.4</td>
+      <td>8924</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w8a8</th>
+      <td>1.06</td>
+      <td>0.9</td>
+      <td>1734</td>
+      <td>4.2</td>
+      <td>8488</td>
+      <td>4.7</td>
+      <td>9548</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.10</td>
+      <td>0.9</td>
+      <td>1775</td>
+      <td>4.2</td>
+      <td>8540</td>
+      <td>5.1</td>
+      <td>10318</td>
+    </tr>
+    <tr>
+      <th rowspan="3" valign="top">H100x1</th>
+      <th>Qwen/Qwen2.5-VL-3B-Instruct</th>
+      <td></td>
+      <td>1.1</td>
+      <td>1188</td>
+      <td>4.3</td>
+      <td>4656</td>
+      <td>4.3</td>
+      <td>4676</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-FP8-Dynamic</th>
+      <td>1.15</td>
+      <td>1.4</td>
+      <td>1570</td>
+      <td>4.3</td>
+      <td>4676</td>
+      <td>4.8</td>
+      <td>5220</td>
+    </tr>
+    <tr>
+      <th>neuralmagic/Qwen2.5-VL-3B-Instruct-quantized.w4a16</th>
+      <td>1.96</td>
+      <td>4.2</td>
+      <td>4598</td>
+      <td>4.1</td>
+      <td>4505</td>
+      <td>4.4</td>
+      <td>4838</td>
+    </tr>
+  </tbody>
+</table>