Pedro Marques ⚙︎
@pmarquees
RL and Evals
💻
🏔️
🚵
🧑🍳
가입 April 2009
1.8K 팔로잉 중 3.5K 팬
Lots of fun new approaches on RL and post training on the timeline today
New research: We developed JOLT, a post-training method where one model acts as both teacher and student, with the teacher trained to give feedback the student can use.
On MATH, it matched GRPO's accuracy with 6.5x fewer training tokens.
더 보기
