Matrix Orthogonalization Improves Memory in Recurrent Modelsayushtambde.com85 points·by at2005·13 hari yang lalu·32 comments
Tree Search Distillation for Language Models Using PPOayushtambde.com87 points·by at2005·4 bulan yang lalu·10 comments
at2005·4 bulan yang lalu·discussI didn't compare with the harness (focused on distillation) but the original ToT paper has a section on it: https://arxiv.org/abs/2305.10601
at2005·4 bulan yang lalu·discussAh, I meant that MCTS uses more inference-time compute (over GRPO) to produce a training sample