peft-rlvr-evaluation

【效率与协作】Comprehensively evaluate 12+ parameter-efficient fine-tuning methods for RL with Verifiable Rewards (RLVR). Show DoRA/AdaLoRA outperform LoRA, SVD-based methods fail on RL, extreme reduction creates bottlenecks—providing empirical evidence that geometric-aware adapters align better with RL's off-principal update dynamics.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准