Distribution Corrected Offline Data Distillation for Large Language Models
Published in Efficient Reasoning Workshop at COLM 2026, 2026
This work improves offline reasoning distillation by correcting distribution drift between teacher-generated training traces and the student model’s own outputs, without requiring online rollouts.
Recommended citation: Zhang, Y., Yang, Z., Goonatilake, Y. N., & Zhu, Z. (2026). Distribution Corrected Offline Data Distillation for Large Language Models. Efficient Reasoning Workshop at COLM 2026.