As the resident “SFT all the things” person at HF, I approve of this message.
Lewis Tunstall backs supervised fine-tuning over expensive reinforcement learning
AILewis Tunstall of Hugging Face says he approves of the message that supervised fine-tuning, or SFT, should be used for everything. A quoted post from Esong Ghori advises climbing as far as possible with SFT before relying on expensive reinforcement learning, or RL. Tunstall adds that RL is the cherry on top, citing Yann LeCun.
Lewis TunstallVerified on X
@_lewtun
One lesson from this launch: climb as far as you can with SFT before relying on expensive RL. RL is still the cherry on top, as @ylecun once put it.
Source details
Lewis Tunstall · x.comPublished · added here
Original title: As the resident “SFT all the things” person at HF, I approve of this message.