Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[NeurIPS 2024 Spotlight] Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models
| Date | Stars |
|---|---|
| 2026-07-31 | 677 |
| 2026-08-05 | 677 |
| 2026-08-06 | 677 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models <a href='https://arxiv.org/abs/2406.04271'><img src='https://img.shields.io/badge/arXiv-2406.04271-b31b1b.svg'></a> Official implementation of our [Buffer of Thoughts (BoT)](https://arxiv.org/abs/2406.04271) framework (**NeurIPS 2024 Spotlight**). Affiliation: Peking University, UC Berkeley, Stanford University ## 📢 News [2025.6] 🎉 Based on our [Buffer of Thoughts (BoT)](https://arxiv.org/abs/2406.04271), we develop [**ReasonFlux-PRM**](https://arxiv.org/abs/2506.18896), a family of trajectory-aware process reward models (PRMs) for long CoT reasoning in LLMs. ReasonFlux-PRM is able to support **both offline and online reward supervision**, by selecting high-quality training data for model distillation, providing dense process-level rewards for policy optimization during reinforcement learning, and enabling reward-guided test-time scaling. Our trained PRMs including [ReasonFlux-PRM-7B](https://huggingface.co/Gen-Verse/ReasonFlux-PRM-7B) and [ReasonFlux-PRM-1.5B](https://huggingface.co/Gen-Verse/ReasonFlux-PRM-1.5B) are now available on [HuggingFace-GenX](https://huggingface.co/Gen-Verse). We also release a 7B advanced thinking and reasoning model [ReasonFlux-PRM-Qwen-2.5-7B](https://huggingface.co/Gen-Verse/ReasonFlux-PRM-Qwen-2.5-7B) supervised via our PRM. [2025.3] 🎉 We release [ReasonFlux-F1-32B](https://huggingface.co/Gen-Verse/ReasonFlux-F1), [ReasonFlux-F1-14B](https://huggingface.co/Gen-Verse/ReasonFlux-F1-14B), [ReasonFlux-F1-7B](https://huggingface.co/Gen-Verse/ReasonFlux-F1-7B), a series of SOTA-level reasoning LLMs by leveraging the **template-augmented reasoning (based on BoT) trajectories** collected from our [ReasonFlux-Zero](https://github.com/Gen-Verse/ReasonFlux). For the training and evaluation scripts, please refer to [Reasonflux-F1](https://github.com/Gen-Verse/ReasonFlux/tree/main/reasonflux-f1) for detail. | Task/Pass@1 | [**ReasonFlux-F1-32B**](https://huggingface.co/Gen-Verse/ReasonFlux-F1) | **ReasonFlux-Zero-32B** | **DeepSeek-R1-Distill-32B** | **o1-mini** | **LIMO -32B** | **s1-32B** | | :------------- | :----------------: | :-------------: | :-------------------: | :-----------------: | :--------: | :--------: | | MATH500 | **96.0** | 91.2 | 94.3 | 90.0 | 90.6 | 93.0 | | AIME 2024 | **76.7** | 56.7 | 72.6 | 56.7 | 50.0 | 56.7 | | AIME 2025 | **53.3** | 37.2 | 46.67 | 50.8 | 37.2 | 49.3 | | GPQA-Diamond | **67.2** | 61.2 | 62.1 | 60.0 | 65.2 | 59.6 | [2024.10] 🎉 We release [SuperCorrect](https://github.com/YangLing0818/SuperCorrect-llm) based on Buffer of Thoughts, a new self-correction LLM reasoning framework. Notably, this new SupperCorrect-7B model achieves SOTA performance on MATH and GSM8K benchmarks among all 7B models. [2024.10] 🎉 We update our repo and release our implementation of **Meta Buffer** and **Buffer Manager** on math problems such as **GSM8K** based on amazing work [light-RAG](https://github.com/HKUDS/LightRAG). [2024.6] 🎉 We release our code for evaluation on three benchmarks. ## Introduction We introduce **BoT**, a novel and versatile thought-augmented reasoning approach designed to enhance the accuracy, efficiency, and robustness of large language models (LLMs). Specifically, we propose a **meta-buffer** to store a series of high-level thoughts, referred to as **thought-templates**, distilled from problem-solving processes across various tasks. For each problem, we retrieve a relevant thought-template and adaptively instantiate it with specific reasoning structures to conduct efficient reasoning. To ensure scalability an
Excerpt of 10,646 characters
Read on GitHub
Ling Yang
22
Zhaochen Yu · National University of Singapore · Singapore
11
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:07d99f3c4b681066, topic:large-language-models
matched fp:07d99f3c4b681066, topic:retrieval-augmented-generation