Update README.md

README.md

@@ -1,3 +1,236 @@
----
-license: gemma
----
+
+---
+tags:
+- vllm
+- vision
+- w4a16
+license: gemma
+base_model: google/gemma-3-27b-it
+library_name: transformers
+---
+
+# gemma-3-27b-it-quantized.w4a16
+
+## Model Overview
+- **Model Architecture:** google/gemma-3-27b-it
+  - **Input:** Vision-Text
+  - **Output:** Text
+- **Model Optimizations:**
+  - **Weight quantization:** INT4
+  - **Activation quantization:** FP16
+- **Release Date:** 6/4/2025
+- **Version:** 1.0
+- **Model Developers:** RedHatAI
+
+Quantized version of [google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights of [google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it) to INT4 data type, ready for inference with vLLM >= 0.8.0.
+
+## 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="nm-testing/gemma-3-27b-it-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:
+
+<details>
+  <summary>Model Creation Code</summary>
+  
+```python
+import base64
+from io import BytesIO
+import torch
+from datasets import load_dataset
+from transformers import AutoProcessor, Gemma3ForConditionalGeneration
+from llmcompressor.modifiers.quantization import GPTQModifier
+from llmcompressor.transformers import oneshot
+
+
+# Load model.
+model_id = "google/gemma-3-27b-it"
+model = Gemma3ForConditionalGeneration.from_pretrained(
+    model_id,
+    device_map="auto",
+    torch_dtype="auto",
+)
+processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
+
+# Oneshot arguments
+DATASET_ID = "neuralmagic/calibration"
+DATASET_SPLIT = {"LLM": "train[:1024]"}
+NUM_CALIBRATION_SAMPLES = 1024
+MAX_SEQUENCE_LENGTH = 2048
+
+# Load dataset and preprocess.
+ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
+ds = ds.shuffle(seed=42)
+
+dampening_frac=0.07
+
+def data_collator(batch):
+    assert len(batch) == 1, "Only batch size of 1 is supported for calibration"
+    item = batch[0]
+    collated = {}
+    import torch
+
+
+    for key, value in item.items():
+        if isinstance(value, torch.Tensor):
+            collated[key] = value.unsqueeze(0)
+        elif isinstance(value, list) and isinstance(value[0][0], int):
+            # Handle tokenized inputs like input_ids, attention_mask
+            collated[key] = torch.tensor(value)
+        elif isinstance(value, list) and isinstance(value[0][0], float):
+            # Handle possible float sequences
+            collated[key] = torch.tensor(value)
+        elif isinstance(value, list) and isinstance(value[0][0], torch.Tensor):
+            # Handle batched image data (e.g., pixel_values as [C, H, W])
+            collated[key] = torch.stack(value)  # -> [1, C, H, W]
+        elif isinstance(value, torch.Tensor):
+            collated[key] = value
+        else:
+            print(f"[WARN] Unrecognized type in collator for key={key}, type={type(value)}")
+    
+    return collated
+   
+
+
+# Recipe
+recipe = [
+    GPTQModifier(
+        targets="Linear",
+        scheme="W4A16",
+        ignore: ["re:.*lm_head.*", "re:.*embed_tokens.*", "re:vision_tower.*", "re:multi_modal_projector.*"],
+        sequential_update: True,
+    )
+]
+
+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 using [lm_evaluation_harness](https://github.com/neuralmagic/lm-evaluation-harness) for OpenLLM v1 text benchmark. The evaluations were conducted using the following commands:
+
+<details>
+<summary>Evaluation Commands</summary>
+
+### OpenLLM v1
+```
+lm_eval \
+  --model vllm \
+  --model_args pretrained="<model_name>",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=<n>,gpu_memory_utilization=0.8,enable_chunked_prefill=True,trust_remote_code=True,enforce_eager=True \
+  --tasks openllm \
+  --batch_size auto
+```
+</details>
+
+
+### Accuracy
+
+<table>
+  <thead>
+    <tr>
+      <th>Category</th>
+      <th>Metric</th>
+      <th>google/gemma-3-27b-it</th>
+      <th>nm-testing/gemma-3-27b-it-quantized.w8a8</th>
+      <th>Recovery (%)</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td rowspan="7"><b>OpenLLM V1</b></td>
+      <td>ARC Challenge</td>
+      <td>72.53%</td>
+      <td>72.35%</td>
+      <td>99.76%</td>
+    </tr>
+    <tr>
+      <td>GSM8K</td>
+      <td>92.12%</td>
+      <td>91.66%</td>
+      <td>99.51%</td>
+    </tr>
+    <tr>
+      <td>Hellaswag</td>
+      <td>85.78%</td>
+      <td>84.97%</td>
+      <td>99.06%</td>
+    </tr>
+    <tr>
+      <td>MMLU</td>
+      <td>77.53%</td>
+      <td>76.77%</td>
+      <td>99.02%</td>
+    </tr>
+    <tr>
+      <td>Truthfulqa (mc2)</td>
+      <td>62.20%</td>
+      <td>62.57%</td>
+      <td>100.59%</td>
+    </tr>
+    <tr>
+      <td>Winogrande</td>
+      <td>79.40%</td>
+      <td>79.79%%</td>
+      <td>100.50%</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>78.26%</b></td>
+      <td><b>78.02%</b></td>
+      <td><b>99.70%</b></td>
+    </tr>
+  </tbody>
+</table>