初始化项目，由ModelHub XC社区提供模型

Model: RedHatAI/pixtral-12b-FP8-dynamic Source: Original Platform
2026-09-20 08:38:12 +08:00
commit 9bd5c221a7
16 changed files with 281 additions and 0 deletions
--- a/README.md
+++ b/README.md
@@ -0,0 +1,151 @@
+---
+tags:
+- fp8
+- vllm
+language:
+- en
+- de
+- fr
+- it
+- pt
+- hi
+- es
+- th
+pipeline_tag: text-generation
+license: apache-2.0
+library_name: vllm
+base_model:
+- mistral-community/pixtral-12b
+- mistralai/Pixtral-12B-2409
+---
+
+# pixtral-12b-FP8-dynamic
+
+## Model Overview
+- **Model Architecture:** Pixtral (Llava)
+  - **Input:** Text/Image
+  - **Output:** Text
+- **Model Optimizations:**
+  - **Weight quantization:** FP8
+  - **Activation quantization:** FP8
+- **Intended Use Cases:** Intended for commercial and research use in multiple languages. Similar to [mistralai/Pixtral-12B-2409](https://huggingface.co/mistralai/Pixtral-12B-2409), this models is intended for assistant-like chat.
+- **Out-of-scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
+- **Release Date:** 11/1/2024
+- **Version:** 1.0
+- **License(s):** [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md)
+- **Model Developers:** Neural Magic
+
+Quantized version of [mistral-community/pixtral-12b](https://huggingface.co/mistral-community/pixtral-12b).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights and activations of [mistral-community/pixtral-12b](https://huggingface.co/mistral-community/pixtral-12b) to FP8 data type, ready for inference with vLLM built from source.
+This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
+
+Only the weights and activations of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the FP8 representations of the quantized weights and activations. Activations are also quantized on a per-token dynamic basis.
+[LLM Compressor](https://github.com/vllm-project/llm-compressor) is used for quantization.
+
+## 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 import LLM, SamplingParams
+
+# Initialize the LLM
+model_name = "neuralmagic/pixtral-12b-FP8-dynamic"
+llm = LLM(model=model_name, max_model_len=10000)
+
+# Create the prompt
+image_url = "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
+messages = [
+    {
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "Describe the image."},
+            {"type": "image_url", "image_url": {"url": image_url}},
+        ],
+    },
+]
+
+# Set up sampling parameters
+sampling_params = SamplingParams(temperature=0.2, max_tokens=100)
+
+# Generate the response
+outputs = llm.chat(messages, sampling_params=sampling_params)
+
+# Print the generated text
+for output in outputs:
+    print(output.outputs[0].text)
+```
+
+vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
+
+```
+vllm serve neuralmagic/pixtral-12b-FP8-dynamic
+```
+
+## Creation
+
+This model was created by applying [LLM Compressor](https://github.com/vllm-project/llm-compressor/blob/f90013702b15bd1690e4e2fe9ed434921b6a6199/examples/quantization_w8a8_fp8/llama3.2_vision_example.py), as presented in the code snipet below.
+
+```python
+from transformers import AutoProcessor, LlavaForConditionalGeneration
+
+from llmcompressor.modifiers.quantization import QuantizationModifier
+from llmcompressor.transformers import oneshot, wrap_hf_model_class
+
+MODEL_ID = "mistral-community/pixtral-12b"
+
+# Load model.
+model_class = wrap_hf_model_class(LlavaForConditionalGeneration)
+model = model_class.from_pretrained(MODEL_ID, device_map="auto", torch_dtype="auto")
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+
+# Configure the quantization algorithm and scheme.
+# In this case, we:
+#   * quantize the weights to fp8 with per channel via ptq
+#   * quantize the activations to fp8 with dynamic per token
+recipe = QuantizationModifier(
+    targets="Linear",
+    scheme="FP8_DYNAMIC",
+    ignore=["re:.*lm_head", "re:multi_modal_projector.*", "re:vision_model.*"],
+)
+
+# Apply quantization and save to disk in compressed-tensors format.
+SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
+oneshot(model=model, recipe=recipe, output_dir=SAVE_DIR)
+processor.save_pretrained(SAVE_DIR)
+
+# Confirm generations of the quantized model look sane.
+print("========== SAMPLE GENERATION ==============")
+input_ids = processor(text="Hello my name is", return_tensors="pt").input_ids.to("cuda")
+output = model.generate(input_ids, max_new_tokens=20)
+print(processor.decode(output[0]))
+print("==========================================")
+```
+
+## Evaluation
+
+### Multimodal Benchmarks
+
+|                   | pixtral-12b | pixtral-12b-FP8-dynamic |
+|:-------------------:|:-------------:|:----------:|
+| **MMMU** *(CoT)*      | 49.44       | 51.11     | 
+| **Mathvista** *(CoT)*   | 58.1        | 59.4     |
+| **ChartQA** *(CoT)*    | 82.64        | 82.68     | 
+| **DocVQA** *(ANLS)*        | 89.36       | 89.35     | 
+
+### Text Benchmarks
+
+|                   | pixtral-12b | pixtral-12b-FP8-dynamic |
+|:-------------------:|:-------------:|:----------:|
+| **MMLU** *(5-shot)*   |  69.27       | 68.96     |
+| **Math** *(0-shot)*         |  43.82       | 43.27     |
+| **Human Eval** *(Pass@1)*    |   77.80      | 76.4     |
+
+### Reproduction
+
+TBD