Tree Search Distillation for Language Models Using PPO
81 points by at2005 20 hours ago | 9 comments
qumpis 6 hours ago
I may never understand what harness means - it's used in so many contexts
replyblamestross 3 hours ago
Its a thing that isn't part of the "subject", used with the subject, to manipulate the state of the "the subject" to be closer to what we want.
replysupermdguy 16 hours ago
> One might note that MCTS uses more inference compute on a per-sample basis than GRPO: of course it performs better
replyThis part confused me, it sounded like they were only doing the MCTS at train time, and then using GRPO to distill the MCTS policy into the model weights. So wouldn’t the model still have the same inference cost?
natufunu 14 hours ago
Great post! I wonder why MCTS is not more popular as a test time compute harness. Did you compare performance of MCTS (without distillation) against other methods (eg best of N) with the same compute budget?
replyrichardvsu 8 hours ago
Why is almost every RL paper done on Qwen-2.5 ? That decreases its credibility.
replyyorwba 6 hours ago
It makes it easier to compare with other papers. If two different papers apply different methods to different models and get different results, how do you know which method is better?
replyOnce you have identified the best method and want to productize it, it would of course make sense to apply it on top of the best model, but if you're just doing research, you can skip that expensive last step.
mapontosevenths 5 hours ago
> Why is almost every RL paper done on Qwen-2.5 ?
replyIn what way does using this model reduce the authors credibility?
algo_trader 11 hours ago
great write up (and effort!! ;))
replywhat are your thoughts on MCTS for coding?
this can/must be paired with a smart execution harness to optimise roll out and roll back of execution paths and system state.
does this change the calculus for optimal post-training ?