Jackwang111/M2RL-RL_Science
Jackwang111/M2RL-RL_Science is a 4 billion parameter language model developed by Haoqing Wang, Xiang Long, Ziheng Li, Yilong Xu, Tingguang Li, and Yehui Tang from Samsung Research and Peking University. This model focuses on multi-domain reinforcement learning for large language models, exploring methods for mixing or merging. It is designed for post-training research, particularly in areas like weight merging and multi-teacher on-policy distillation, and has a context length of 32768 tokens.
Loading preview...
Overview
Jackwang111/M2RL-RL_Science is a 4 billion parameter language model developed by researchers from Samsung Research and Peking University. This model is a result of the paper "To Mix or To Merge? Toward Multi-Domain Reinforcement Learning for Large Language Models," which was accepted to COLM 2026. It explores advanced techniques in multi-domain reinforcement learning for large language models, specifically focusing on strategies for combining or integrating different model components.
Key Capabilities
- Multi-Domain Reinforcement Learning: Investigates methods for effective reinforcement learning across multiple domains for LLMs.
- Post-Training Research: Designed to support further research in areas such as weight merging and multi-teacher on-policy distillation.
- High Context Length: Features a substantial context window of 32768 tokens, enabling processing of longer sequences.
Good For
- Academic Research: Ideal for researchers exploring novel approaches in multi-domain reinforcement learning for LLMs.
- Model Merging Experiments: Provides a foundation for experiments involving the merging of model weights.
- Distillation Techniques: Useful for developing and testing multi-teacher on-policy distillation methods.