New Method Aligns Action Sequences for Multi-Turn On-Policy Distillation
A new arXiv paper proposes action-sequence alignment for multi-turn on-policy distillation, supervising a student model on its own trajectories to reduce train-inference distribution mismatch in tool-use settings.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-11 05:00 · arXiv cs.LG
Which Successes Make Good References? Action-Sequence Alignment for Multi-Turn On-Policy Distillation