treegrpo-tree-advantage-rl

【效率与协作】Achieve 2.4× faster RL training for diffusion models by restructuring denoising as tree search with shared computation. TreeGRPO computes fine-grained step advantages instead of trajectory-level rewards—crucial for efficient diffusion model optimization.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准