Improve SuperCorrect-7B model card

README.md

@@ -1,16 +1,18 @@
 ---
-license: apache-2.0
+base_model:
+- Qwen/Qwen2.5-Math-7B-Instruct
 language:
 - en
+library_name: transformers
+license: apache-2.0
 metrics:
 - accuracy
-base_model:
-- Qwen/Qwen2.5-Math-7B-Instruct
-library_name: transformers
+pipeline_tag: question-answering
 ---
+
 ## SuperCorrect: Supervising and Correcting Language Models with Error-Driven Insights
 
-> [SuperCorrect: Supervising and Correcting Language Models with Error-Driven Insights](link)
+> [SuperCorrect: Supervising and Correcting Language Models with Error-Driven Insights](https://arxiv.org/abs/2410.09008)
 > [Ling Yang\*](https://yangling0818.github.io/), [Zhaochen Yu*](https://github.com/BitCodingWalkin), [Tianjun Zhang](https://tianjunz.github.io/), [Minkai Xu](https://minkaixu.com/), [Joseph E. Gonzalez](https://people.eecs.berkeley.edu/~jegonzal/),[Bin Cui](https://cuibinpku.github.io/), [Shuicheng Yan](https://yanshuicheng.info/)
 >
 > Peking University, Skywork AI, UC Berkeley, Stanford University 
@@ -20,46 +22,23 @@ library_name: transformers
   <img src='https://img.shields.io/badge/Arxiv-2410.09008-A42C25?style=flat&logo=arXiv&logoColor=A42C25'></a> 
 </p>
 
-## Introduction
-
-![image](intro.png)
+## Abstract
 
-This repo provides the official implementation of **SuperCorrect**  a novel two-stage fine-tuning method for improving both reasoning accuracy and self-correction ability for LLMs. 
+Large language models (LLMs) like GPT-4, PaLM, and LLaMA have shown significant improvements in various reasoning tasks. However, smaller models such as Llama-3-8B and DeepSeekMath-Base still struggle with complex mathematical reasoning because they fail to effectively identify and correct reasoning errors. Recent reflection-based methods aim to address these issues by enabling self-reflection and self-correction, but they still face challenges in independently detecting errors in their reasoning steps. To overcome these limitations, we propose SuperCorrect, a novel two-stage framework that uses a large teacher model to supervise and correct both the reasoning and reflection processes of a smaller student model. In the first stage, we extract hierarchical high-level and detailed thought templates from the teacher model to guide the student model in eliciting more fine-grained reasoning thoughts. In the second stage, we introduce cross-model collaborative direct preference optimization (DPO) to enhance the self-correction abilities of the student model by following the teacher's correction traces during training. This cross-model DPO approach teaches the student model to effectively locate and resolve erroneous thoughts with error-driven insights from the teacher model, breaking the bottleneck of its thoughts and acquiring new skills and knowledge to tackle challenging problems. Extensive experiments consistently demonstrate our superiority over previous methods. Notably, our SuperCorrect-7B model significantly surpasses powerful DeepSeekMath-7B by 7.8%/5.3% and Qwen2.5-Math-7B by 15.1%/6.3% on MATH/GSM8K benchmarks, achieving new SOTA performance among all 7B models. Code: https://github.com/YangLing0818/SuperCorrect-llm
 
-Notably, our **SupperCorrect-7B** model significantly surpasses powerful **DeepSeekMath-7B by 7.8%/5.3% and Qwen2.5-Math-7B by 15.1%/6.3% on MATH/GSM8K benchmarks**, achieving new SOTA performance among all 7B models.
 
-<div align="left">
-🚨 Unlike other LLMs, we incorporate LLMs with our pre-defined hierarchical thought template ([Buffer of Thought (BoT)](https://github.com/YangLing0818/buffer-of-thought-llm)) to conduct more deliberate reasoning than conventional CoT. It should be noted that our evaluation methods relies on pure mathematical reasoning abilities of LLMs, instead of leverage other programming methods such as PoT and ToRA.
-</div>
-
-## Examples
+## Introduction
 
-![image](example1.png)
+![image](imgs/intro.png)
 
-<div align="left">
-<b>
-🚨 For more concise and clear presentation, we omit some XML tags. 
-</b>
-</div>
+This repo provides the official implementation of **SuperCorrect**,  a novel two-stage fine-tuning method for improving both reasoning accuracy and self-correction ability for LLMs. We incorporate LLMs with our pre-defined hierarchical thought template ([Buffer of Thought (BoT)](https://github.com/YangLing0818/buffer-of-thought-llm)) to conduct more deliberate reasoning than conventional CoT. 
 
-### Model details
-*You can check our [Github repo](https://github.com/YangLing0818/SuperCorrect-llm) for more details.*
 
 ## Quick Start
 
-### Requirements
-
-* Since our current model is  based on Qwen2.5-Math series, `transformers>=4.37.0` is needed for Qwen2.5-Math models. The latest version is recommended.
-
-> [!Warning]
->
-> <div align="center">
-> <b>
-> 🚨 This is a must because `transformers` integrated Qwen2 codes since `4.37.0`.
-> </b>
-> </div>
+**(See the Github README for more detailed installation and usage instructions.)**
 
-### Inference
+### Inference with Different Libraries
 
 #### 🤗 Hugging Face Transformers
 
@@ -103,29 +82,34 @@ response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
 print(response)
 ```
 
-## Performance
+#### 🔥 vLLM
 
-We evaluate our SupperCorrect-7B on two widely used English math benchmarks GSM8K and MATH. All evaluations are tested with our evaluation method which is zero-shot hierarchical thought based prompting.
+```python
+# (See Github README for vLLM inference example)
+```
+
+## Evaluation
 
-![image](table.png)
+**(See the Github README for the most up-to-date evaluation results and instructions.)**
 
 ## Citation
 
 ```bash
-@article{yang2024supercorrect,
-title={SuperCorrect: Supervising and Correcting Language Models with Error-Driven Insights}
+@inproceedings{yang2025supercorrect,
+  title={SuperCorrect: Supervising and Correcting Language Models with Error-Driven Insights},
   author={Yang, Ling and Yu, Zhaochen and Zhang, Tianjun and Xu, Minkai and Gonzalez, Joseph E and Cui, Bin and Yan, Shuicheng},
-  journal={arXiv preprint arXiv:2410.09008},
-  year={2024}
+  booktitle={International Conference on Learning Representations},
+  year={2025}
 }
+
 @article{yang2024buffer,
   title={Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models},
   author={Yang, Ling and Yu, Zhaochen and Zhang, Tianjun and Cao, Shiyi and Xu, Minkai and Zhang, Wentao and Gonzalez, Joseph E and Cui, Bin},
-  journal={arXiv preprint arXiv:2406.04271},
+  journal={Advances in Neural Information Processing Systems},
   year={2024}
 }
 ```
 
 ## Acknowledgements
 
-Our SuperCorrect is a two-stage fine-tuning model which based on several extraordinary open-source models like [Qwen2.5-Math](https://github.com/QwenLM/Qwen2.5-Math), [DeepSeek-Math](https://github.com/deepseek-ai/DeepSeek-Math), [Llama3-Series](https://github.com/meta-llama/llama3). Our evaluation method is based on the code base of outstanding works like [Qwen2.5-Math](https://github.com/QwenLM/Qwen2.5-Math) and  [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness). We also want to express our gratitude for amazing works such as [BoT](https://github.com/YangLing0818/buffer-of-thought-llm) which provides the idea of thought template.
+Our SuperCorrect is a two-stage fine-tuning model based on several extraordinary open-source models like [Qwen2.5-Math](https://github.com/QwenLM/Qwen2.5-Math), [DeepSeek-Math](https://github.com/deepseek-ai/DeepSeek-Math), [Llama3-Series](https://github.com/meta-llama/llama3). Our evaluation method is based on the code base of outstanding works like [Qwen2.5-Math](https://github.com/QwenLM/Qwen2.5-Math) and  [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness). We also want to express our gratitude for amazing works such as [BoT](https://github.com/YangLing0818/buffer-of-thought-llm) which provides the idea of thought template.