Official research code for the paper Rethinking On-Policy Distillation of Large Language Models. It provides OPD, SFT and GRPO training recipes, evaluation scripts, released checkpoints and diagnostics for reproducing the reported findings.
1072llms,mechanism,on-policy-distillationOfficial research code for the paper Rethinking On-Policy Distillation of Large Language Models. It provides OPD, SFT and GRPO training recipes, evaluation scripts, released checkpoints and diagnostics for reproducing the reported findings.