Press Space to continue
Supervised Fine-Tuning Can Outperform Reinforcement Learning
Press Space to continue
"Finetuning with Sampling: SFT Learns Better Than You Think" SFT may be better than we think, and the problem could have just been the data distribution. This paper uses MCMC to rewrite expert traces into equally correct answers that look more like what the base model would naturally produce. With that, plain SFT can outperform RL while forgetting less! https://alphaxiv.org/abs/2610.02140
open on x · 3 viewsViews and clicks count once per visitor per day. Clicks are opens of this post on X.