Update README.md

README.md

@@ -24,6 +24,7 @@ pipeline_tag: text-generation
 ## Overview
 QwQ‑32B‑Distill‑Qwen‑1.5B‑Alpha is a groundbreaking language model built on top of the DeepSeek‑R1‑Distill‑Qwen‑1.5B base. Developed entirely by a solo innovator—with valuable inspiration from Berkeley’s research—the model employs a novel reinforcement learning distillation framework that dramatically enhances performance while keeping training data requirements and compute costs to a minimum. Despite having only 1.5B parameters, the model achieves a striking 47.18 MMLU score and outperforms prior baselines on multiple math and reasoning benchmarks.
 
+---
 
 ## Data
 Our training dataset is comprised of 6,170 meticulously curated problem–answer pairs drawn from high-quality sources such as:
@@ -34,6 +35,8 @@ Our training dataset is comprised of 6,170 meticulously curated problem–answer
 
 By focusing on a lean yet highly informative dataset, the model efficiently learns critical reasoning capabilities without the burden of excessive data volume.
 
+---
+
 ## Training Recipe
 To maximize performance with minimal resources, QwQ‑32B‑Distill‑Qwen‑1.5B‑Alpha utilizes an innovative training strategy that includes:
 
@@ -46,6 +49,7 @@ Progressive expansion of the context length is used to boost model performance w
 
 Training was carried out using <b>H200 GPUs for 336 hours</b> at an exceptionally low cost of approximately <b>$1,341</b>. This carefully engineered approach makes it possible to obtain state-of-the-art performance with very limited training data.
 
+---
 
 ## Evaluation
 The model has been rigorously evaluated on a variety of challenging benchmarks. Below is a snapshot of the results:
@@ -59,6 +63,8 @@ The model has been rigorously evaluated on a variety of challenging benchmarks.
 | **AMC 2023**     | 75.00               | 58.92                | 44.17                |
 | **MATH 5000**    | 38.89               | –                    | 20,173               |
 
+
+### Comparison
 | Model                                 | Base Model                             | Parameter Num | Training Time | GPU Amount (Cloud GPU) | Training Data量 | MMLU | AIME 2024 | AIME 2025-I | AIME 2025-II | AMC 2023 | MATH 5000 | Average |
 |--------------------------------------|----------------------------------------|----------------|----------------|--------------------------|------------------|------|------------|--------------|---------------|-----------|-------------|---------|
 |                                      |                                        |                |                |                          |                  | Path@1 | Cons@64 | Avg. Token Count | Path@1 | Cons@64 | Avg. Token Count | Path@1 | Cons@64 | Avg. Token Count | Path@1 | Path@1 | Path@1 | Cons@64 | Token Count |
@@ -66,3 +72,30 @@ The model has been rigorously evaluated on a variety of challenging benchmarks.
 | DeepScaleR-1.5B-Preview              | DeepSeek-R1-Distill-Qwen-1.5B          | 1.5B           | A100*3600H     | $6,804                   | 40,000           | 36.15 | <strgin>39.37</strong> | <strong>56.67</strong> | 29,888 | <strong>36.77</strong> | <strong>53.33</strong> | 33,151 | 20.00 | 20.00 | 35,293 | <strong>80.00</strong> | <strong>61.96</strong> | <strong>45.87</strong> | <strong>43.33</strong> | 32,777 |
 | <strong>QwQ-32B-Distill-Qwen-1.5B-Alpha</strong> | DeepSeek-R1-Distill-Qwen-1.5B| 1.5B | <strong>H200*336H</strong> | <strong>$1,341</strong> | <strong>6,170</strong> | <strong>47.18</strong> | 27.76 | 43.33 | <strong>21,191</strong> | 34.58 | 40.00 | <strong>17,952</strong> | <strong>21.56</strong> | <strong>33.33</strong> | <strong>21,376</strong> | 75.00 | 58.92 | 44.17 | 38.89 | <strong>20,173</strong> |
 | Phi-4-mini-instruct                  | -                                      | 3B             | -              | -                        | -                | 13.79 | 13.79 | -     | 2,055  | 20.00 | 20.00 | 2,001  | 0.00  | 0.00  | 1,970  | -     | -     | 11.26 | 11.26 | 2,009  |
+
+---
+
+## Serving QwQ‑32B‑Distill‑Qwen‑1.5B‑Alpha
+
+Deploy your model effortlessly using high-performance inference systems, including:
+
+- **vLLM**  
+- **Hugging Face Text Generation Inference (TGI)**  
+- **SGLang**  
+- **TensorRT-LLM**
+
+All these systems support the OpenAI Chat Completions API format, ensuring smooth integration into your applications.
+
+
+---
+
+## License
+
+This project is released under the **MIT License**, reflecting our commitment to open and accessible AI. We firmly believe that cutting-edge AI research should be available for anyone to use, modify, and build upon.
+
+
+---
+
+## Acknowledgements
+
+A huge thank you is due to the open-source community and the team behind **DeepSeek‑R1‑Distill‑Qwen‑1.5B**, whose groundwork made this innovation possible. This model is a testament to what can be achieved with passion, creativity, and a singular vision—developed entirely by a solo researcher while drawing inspiration from Berkeley’s forward-thinking research ecosystem.