kapynResearch

Can GRPO be 10x Efficient? Kwai AI’s SRPO Suggests Yes with SRPO

Kwai AI's SRPO framework drastically reduces LLM RL post-training steps by 90%. This novel two-stage reinforcement learning approach with history resampling achieves performance on par with DeepSeek-R1 in math and code benchmarks, suggesting a path to significantly more efficient LLM fine-tuning.

Synced·Apr 24, 2025

Opening Kapyn…