We need to produce HTML summary with

and link. Use only provided info. Title: TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training. Description mentions OPD, inefficiencies: full-horizon rollouts waste resources on tail turns, trajectory-level KL objectives concentrate most of... cut off. We can summarize: TurnOPD improves on-policy distillation by making it turn-aware to reduce waste and focus supervision on informative turns for long-horizon tasks. Provide link. We must output only HTML, no extra text. Provide

...

then Read original. Ensure correct quoting. Let's craft 3 sentences. Sentence1: TurnOPD enhances on-policy distillation by introducing turn-awareness to focus supervision on informative steps in long-horizon agent tasks. Sentence2: It addresses the inefficiency of full-horizon rollouts that waste compute on tail turns with weak KL signals. Sentence3: By reweighting the KL objective per turn, TurnOPD reduces wall-clock cost while preserving learning effectiveness. That's 3 sentences. Now HTML:

TurnOPD enhances on-policy distillation by introducing turn-awareness to focus supervision on informative steps in long-horizon agent tasks. It addresses the inefficiency of full-horizon rollouts that waste compute