Sampling Trick Makes Simple Fine-Tuning Match Reinforcement Learning

Original: Finetuning with Sampling: SFT Learns Better Than You Think

Source: arxiv.org ↗