Matrix Orthogonalization Improves Memory in Recurrent Modelsayushtambde.com85 points·by at2005·13 dni temu·32 comments
Tree Search Distillation for Language Models Using PPOayushtambde.com87 points·by at2005·4 miesiące temu·10 comments
at2005·4 miesiące temu·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 miesiące temu·discussAh, I meant that MCTS uses more inference-time compute (over GRPO) to produce a training sample