| language: | |
| - en | |
| datasets: | |
| - qingfei1/R-Search_datasets | |
| license: apache-2.0 | |
| # R-Search: Empowering LLM Reasoning with Search via Multi-Reward Reinforcement Learning | |
| <p align="center"> | |
| 🤗 <a href="https://huggingface.co/datasets/qingfei1/R-Search_datasets" target="_blank">[R-Search Datasets] </a> • 💻 <a href="https://github.com/QingFei1/R-Search" target="_blank">[Github Repo]</a> | |
| </p> | |
| **R-Search** is a novel reinforcement learning framework for reasoning–search integration. It enables LLMs to autonomously perform multi-step reasoning with deep search interaction, and to learn optimal reasoning–search trajectories via multi-reward signals, substantially improving performance on complex logic- and knowledge-intensive tasks. | |
| ## Trained Models | |
| We open-sourced the following models trained only on the 2wikimultihopqa training set: | |
| |Model|Huggingface Repo|Description| | |
| |---|---|---| | |
| |**R-Search-7b-grpo**| [🤗 Huggingface Repo](https://huggingface.co/qingfei1/R-Search-7b-grpo) | Trained **Qwen2.5-7B-Instruct** using the GRPO algorithm | | |
| |**R-Search-3b-grpo**| [🤗 Huggingface Repo](https://huggingface.co/qingfei1/R-Search-3b-grpo) | Trained **Qwen2.5-3B-Instruct** using the GRPO algorithm | | |
| |**R-Search-7b-ppo**| [🤗 Huggingface Repo](https://huggingface.co/qingfei1/R-Search-7b-ppo) | Trained **Qwen2.5-7B-Instruct** using the PPO algorithm | | |
| |**R-Search-3b-ppo**| [🤗 Huggingface Repo](https://huggingface.co/qingfei1/R-Search-3b-ppo) | Trained **Qwen2.5-3B-Instruct** using the PPO algorithm | |