作者位次 #7/9ACL 2026Findings· Findings of ACL
Fast and Effective On-Policy Distillation from Reasoning Prefixes
Dongxu Zhang · Zhichao Yang · Sepehr Janghorbani · Jun Han · Andrew Ressler II · Qian Qian · Gregory D Lyng · Sanjit Singh Batra · Robert E. Tillman
摘要
On-policy distillation (OPD), which samples trajectories from the student model and supervises them with a teacher at the token-level, avoids relying solely on verifiable terminal rewards and can yield better generalization than off-policy distillation. However, OPD requires expensive on-the-fly...