Pinned Loading

1. THUDM/slimeTHUDM/slimePublic

   slime is an LLM post-training framework for RL Scaling.

   Python 8.2k 1.2k

2. OpenRLHF/OpenRLHFOpenRLHF/OpenRLHFPublic

   An Easy-to-use, Scalable and High-performance Agentic RL Framework based on Ray (PPO & DAPO & REINFORCE++ & VLM & TIS & vLLM & Ray & Async RL)

   Python 10k 1k

3. THUDM/SWE-DevTHUDM/SWE-DevPublic

   [ACL25' Findings] SWE-Dev is an SWE agent with a scalable test case construction pipeline.

   Python 65

4. Generalization-of-TransformersGeneralization-of-TransformersPublic

   [ICLR'25] Understanding the Generalization of In-Context Learning in Transformers: An Empirical Study

   Python 4