LoRA tuning behaves differently in fine-tuning versus reinforcement learning

Original: Spectral and Subspace Dynamics of LoRA in Supervised Fine-Tuning and Reinforcement Learning

Source: arxiv.org ↗