Add Mix-Quant paper, project page, and code links

README.md

@@ -1,7 +1,5 @@
 ---
-tags:
-- fp4
-- vllm
+base_model: Qwen/Qwen3-8B
 language:
 - en
 - de
@@ -11,13 +9,22 @@ language:
 - hi
 - es
 - th
-pipeline_tag: text-generation
 license: apache-2.0
-base_model: Qwen/Qwen3-8B
+pipeline_tag: text-generation
+library_name: vllm
+tags:
+- fp4
+- vllm
+- mix-quant
 ---
 
 # Qwen3-8B-NVFP4
 
+This repository contains a quantized version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) as presented in the paper [Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMs](https://huggingface.co/papers/2605.20315).
+
+- **Project Page:** [https://haiquanlu.github.io/Mix-Quant/](https://haiquanlu.github.io/Mix-Quant/)
+- **Code:** [https://github.com/haiquanlu/Mix-Quant](https://github.com/haiquanlu/Mix-Quant)
+
 ## Model Overview
 - **Model Architecture:** Qwen/Qwen3-8B
   - **Input:** Text
@@ -30,12 +37,11 @@ base_model: Qwen/Qwen3-8B
 - **Version:** 1.0
 - **Model Developers:** RedHatAI
 
-This model is a quantized version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).
-It was evaluated on a several tasks to assess the its quality in comparison to the unquatized model.
+This model was evaluated on several tasks to assess its quality in comparison to the unquantized model.
 
 ### Model Optimizations
 
-This model was obtained by quantizing the weights and activations of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) to FP4 data type, ready for inference with vLLM>=0.9.1
+This model was obtained by quantizing the weights and activations of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) to FP4 data type, ready for inference with vLLM>=0.9.1.
 This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%.
 
 Only the weights and activations of the linear operators within transformers blocks are quantized using [LLM Compressor](https://github.com/vllm-project/llm-compressor).
@@ -72,11 +78,11 @@ generated_text = outputs[0].outputs[0].text
 print(generated_text)
 ```
 
-vLLM aslo supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
+vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
 
 ## Creation
 
-This model was created by applying [LLM Compressor with calibration samples from UltraChat](https://github.com/vllm-project/llm-compressor/blob/main/examples/quantization_w4a4_fp4/llama3_example.py), as presented in the code snipet below.
+This model was created by applying [LLM Compressor with calibration samples from UltraChat](https://github.com/vllm-project/llm-compressor/blob/main/examples/quantization_w4a4_fp4/llama3_example.py), as presented in the code snippet below.
 
 <details>
   
@@ -177,13 +183,17 @@ oneshot(
     output_dir=SAVE_DIR,
 )
 
-print("\n\n")
+print("
+
+")
 print("========== SAMPLE GENERATION ==============")
 dispatch_for_generation(model)
 input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to("cuda")
 output = model.generate(input_ids, max_new_tokens=100)
 print(tokenizer.decode(output[0]))
-print("==========================================\n\n")
+print("==========================================
+
+")
 
 model.save_pretrained(SAVE_DIR, save_compressed=True)
 tokenizer.save_pretrained(SAVE_DIR)
@@ -192,7 +202,7 @@ tokenizer.save_pretrained(SAVE_DIR)
 
 ## Evaluation
 
-This model was evaluated on the well-known OpenLLM v1, OpenLLM v2 and HumanEval_64 benchmarks using [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness). The Reasoning evals were done using [ligheval](https://github.com/neuralmagic/lighteval).
+This model was evaluated on the well-known OpenLLM v1, OpenLLM v2 and HumanEval_64 benchmarks using [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness). The Reasoning evals were done using [lighteval](https://github.com/neuralmagic/lighteval).
 
 ### Accuracy
 <table>
@@ -327,9 +337,6 @@ This model was evaluated on the well-known OpenLLM v1, OpenLLM v2 and HumanEval_
   </tbody>
 </table>
 
-
-
-
 ### Reproduction
 
 The results were obtained using the following commands:
@@ -346,7 +353,6 @@ lm_eval \
   --batch_size auto
 ```
 
-
 #### OpenLLM v2
 ```
 lm_eval \
@@ -392,6 +398,16 @@ lighteval vllm model_args.yaml \
   "lighteval|aime24|0,lighteval|aime25|0,lighteval|gpqa:diamond|0" \
   --max-samples -1 \
   --output-dir out_dir
-
 ```
-</details>
+</details>
+
+## Citation
+
+```bibtex
+@article{lu2026mixquant,
+  title={Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMs},
+  author={Lu, Haiquan and Chen, Zigeng and Fang, Gongfan and Ma, Xinyin and Wang, Xinchao},
+  journal={arXiv preprint arXiv:2605.20315},
+  year={2026}
+}
+```