Update README.md

README.md

@@ -1,5 +1,80 @@
+---
+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-VL-72B-Instruct
+library_name: transformers
+---
+
+# Qwen2-VL-72B-Instruct-quantized-w4a16
+
+## Model Overview
+- **Model Architecture:** Qwen/Qwen2-VL-72B-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-VL-72B-Instruct](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights of [Qwen/Qwen2-VL-72B-Instruct](https://huggingface.co/Qwen/Qwen2-VL-72B-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-VL-72B-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
 from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
 
@@ -34,4 +109,223 @@ input_ids = processor(text="Hello my name is", return_tensors="pt").input_ids.to
 output = model.generate(input_ids, max_new_tokens=20)
 print(processor.decode(output[0]))
 print("==========================================")
-```
+```
+</details>
+
+## Evaluation
+
+
+<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-VL-72B-Instruct-FP8-Dynamic --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>
+    <tr>
+      <td>A100x4</td>
+      <td>Qwen/Qwen2-VL-72B-Instruct</td>
+      <td></td>
+      <td>6.5</td>
+      <td>77</td>
+      <td>4.6</td>
+      <td>110</td>
+      <td>4.4</td>
+      <td>113</td>
+    </tr>
+    <tr>
+      <td>A100x2</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w8a8</td>
+      <td>1.85</td>
+      <td>7.2</td>
+      <td>139</td>
+      <td>4.9</td>
+      <td>206</td>
+      <td>4.8</td>
+      <td>211</td>
+    </tr>
+    <tr>
+      <td>A100x1</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w4a16</td>
+      <td>3.32</td>
+      <td>10.0</td>
+      <td>202</td>
+      <td>5.0</td>
+      <td>398</td>
+      <td>4.8</td>
+      <td>419</td>
+    </tr>
+    <tr>
+      <td>H100x4</td>
+      <td>Qwen/Qwen2-VL-72B-Instruct</td>
+      <td></td>
+      <td>4.4</td>
+      <td>66</td>
+      <td>3.0</td>
+      <td>97</td>
+      <td>2.9</td>
+      <td>99</td>
+    </tr>
+    <tr>
+      <td>H100x2</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-FP8-Dynamic</td>
+      <td>1.79</td>
+      <td>4.7</td>
+      <td>119</td>
+      <td>3.3</td>
+      <td>173</td>
+      <td>3.2</td>
+      <td>177</td>
+    </tr>
+    <tr>
+      <td>H100x1</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w4a16</td>
+      <td>2.60</td>
+      <td>6.4</td>
+      <td>172</td>
+      <td>4.3</td>
+      <td>253</td>
+      <td>4.2</td>
+      <td>259</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>
+    <tr>
+      <th rowspan="3" valign="top">A100x4</th>
+      <td>Qwen/Qwen2-VL-72B-Instruct</td>
+      <td></td>
+      <td>0.3</td>
+      <td>169</td>
+      <td>1.1</td>
+      <td>538</td>
+      <td>1.2</td>
+      <td>595</td>
+    </tr>
+    <tr>
+      <td>A100x2</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w8a8</td>
+      <td>1.84</td>
+      <td>0.6</td>
+      <td>293</td>
+      <td>2.0</td>
+      <td>1021</td>
+      <td>2.3</td>
+      <td>1135</td>
+    </tr>
+    <tr>
+      <td>A100x1</td>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w4a16</td>
+      <td>2.73</td>
+      <td>0.6</td>
+      <td>314</td>
+      <td>3.2</td>
+      <td>1591</td>
+      <td>4.0</td>
+      <td>2019</td>
+    </tr>
+    <tr>
+      <td>H100x4</td>
+      <td>Qwen/Qwen2-VL-72B-Instruct</td>
+      <td></td>
+      <td>0.5</td>
+      <td>137</td>
+      <td>1.2</td>
+      <td>356</td>
+      <td>1.3</td>
+      <td>377</td>
+    </tr>
+    <tr>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-FP8-Dynamic</td>
+      <td>H100x2</td>
+      <td>1.70</td>
+      <td>0.8</td>
+      <td>236</td>
+      <td>2.2</td>
+      <td>623</td>
+      <td>2.4</td>
+      <td>669</td>
+    </tr>
+    <tr>
+      <td>neuralmagic/Qwen2-VL-72B-Instruct-quantized.w4a16</td>
+      <td>H100x1</td>
+      <td>2.35</td>
+      <td>1.3</td>
+      <td>350</td>
+      <td>3.3</td>
+      <td>910</td>
+      <td>3.6</td>
+      <td>994</td>
+    </tr>
+  </tbody>
+</table>
+