Update README.md

README.md

@@ -11,50 +11,80 @@ model-index:
   results: []
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
+# ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates
+Revolutionary inference-scaling paradigm with a hierarchical RL algorithm: enpowering a 32B model with 500 thought templates to outperform o1-preview and DeepSeek-V3 in reasoning tasks.
+| Task           | ReasonFlux-F1-32B | **ReasonFlux -Zero-32B** | **R1-Distill-32B** | **OpenAI o1-mini** | **LIMO -32B** | **s1-32B** |
+| :------------- | :----------------: | :-------------: | :-------------------: | :-----------------: | :--------: | :--------: |
+| MATH500           |      **96.0**      |      91.2      |      94.3      |        90.0        |        90.6         |    84.8    |
+| AIME 2024      |      **76.67**      |      56.7      |      72.6      |        56.7        |        50.0         |    36.0    |
+| AIME 2025    | **53.33**         | 37.2                     |        46.67        |         50.8         |        37.2         |    26.7    |
+| GPQA-Diamond | **67.17**         | 61.2                     |      62.1      |        60.0        |        65.2         |    59.6    |
 
 # ReasonFlux-F1-32B
 
-This model is a fine-tuned version of [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) on the ReasonFlux-F1-SFT dataset.
-
-## Model description
-
-More information needed
-
-## Intended uses & limitations
-
-More information needed
-
-## Training and evaluation data
-
-More information needed
-
-## Training procedure
-
-### Training hyperparameters
-
-The following hyperparameters were used during training:
-- learning_rate: 1e-05
-- train_batch_size: 1
-- eval_batch_size: 8
-- seed: 42
-- distributed_type: multi-GPU
-- num_devices: 8
-- gradient_accumulation_steps: 2
-- total_train_batch_size: 16
-- total_eval_batch_size: 64
-- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
-- lr_scheduler_type: cosine
-- num_epochs: 5.0
-
-### Training results
-
-
-
-### Framework versions
-
-- Transformers 4.49.0
-- Pytorch 2.5.1+cu124
-- Datasets 3.2.0
-- Tokenizers 0.21.0
+> ReasonFlux-F1-32B is our finetuned SOTA-level reasoning LLMs by leveraging the template-augmented reasoning trajectories based on our ReasonFlux-Zero. 
+
+* Github Repository: [Gen-Verse/ReasonFlux](https://github.com/Gen-Verse/ReasonFlux)
+* Paper:[ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates](https://arxiv.org/abs/2502.06772)
+* Dataset: [Gen-Verse/ReasonFlux-F1-SFT](https://huggingface.co/datasets/Gen-Verse/ReasonFlux-F1-SFT)
+* Base Model: [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B)
+
+
+## Evaluation
+We present the evaluation results of our ReasonFlux-F1-32B on challenging reasoning tasks including AIME2024,AIM2025,MATH500 and GPQA-Diamond. To make a fair comparison, we report the results of the LLMs on our evaluation scripts in [ReasonFlux-F1]().
+
+| Model                                   | AIME2024@pass1 | AIME2025@pass1 | MATH500@pass1 | GPQA@pass1 |
+| --------------------------------------- | -------------- | -------------- | ------------- | ---------- |
+| QwQ-32B-Preview                         | 46.7           | 37.2           | 90.6          | 65.2       |
+| LIMO-32B                                | 56.3           | 44.5           | 94.80         | 58.08      |
+| s1-32B                                  | 56.7           | 26.7           | 93.0          | 59.6       |
+| OpenThinker-32B                         | 66.0           | 53.3           | 94.8          | 60.10      |
+| FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview | 76.67          | 40.0           | 93.4          | 59.09      |
+| R1-Distill-32B                          | 70             | 46.67          | 92            | 59.59      |
+| ReasonFlux-Zero-32B                     | 56.7           | 37.2           | 91.2          | 61.2       |
+| **ReasonFlux-F1-32B**                   | **76.67**      | **53.33**      | **96.0**      | **67.17**  |
+
+
+## Quick start with VLLM
+```python
+from vllm import LLM, SamplingParams
+from transformers import AutoTokenizer
+
+model_id = 'Gen-Verse/ReasonFlux-F1'
+
+model = LLM(
+    model_id,
+    tensor_parallel_size=8,
+)
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+# stop_token_ids = tok("<|im_end|>\n")["input_ids"]
+
+sampling_params = SamplingParams(
+    max_tokens=32768,
+)
+# 2022 AIME I Problems/Problem 15
+question = """Let \(x, y\), and \(z\) be positive real numbers satisfying the system of equations:
+\[
+\begin{array}{c}
+\sqrt{2 x-x y}+\sqrt{2 y-x y}=1 \\
+\sqrt{2 y-y z}+\sqrt{2 z-y z}=\sqrt{2} \\
+\sqrt{2 z-z x}+\sqrt{2 x-z x}=\sqrt{3} .
+\end{array}
+\]
+Then \(\left[(1-x)(1-y)(1-z)\right]^{2}\) can be written as \(\frac{m}{n}\), where \(m\) and \(n\) are relatively prime positive integers. Find \(m+n\)."""
+ds_prompt="<｜User｜>\n" + question + "<｜Assistant｜>\n"
+output = model.generate(ds_prompt, sampling_params=sampling_params)
+print(output[0].outputs[0].text)
+
+```
+## Citation
+
+```bash
+@article{yang2025reasonflux,
+  title={ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates},
+  author={Yang, Ling and Yu, Zhaochen and Cui, Bin and Wang, Mengdi},
+  journal={arXiv preprint arXiv:2502.06772},
+  year={2025}
+}
+```