Papers

Filtered to on-policy distillation · clear filter

Browse by term

continual learning 89reinforcement learning 43vision-language models 19benchmarking 12benchmarks 12robotics 12language models 9world models 9large language models 8video generation 7multimodal models 5recursive self-improvement 5video world models 5agents 4diffusion models 4embodied AI 4foundation models 4generative models 4human-computer interaction 4multimodal large language models 4natural language processing 4policy optimization 4reasoning 4self-supervised learning 4spatial reasoning 4agentic models 3conversational AI 3dialogue systems 3evaluation benchmarks 3LLMs 3

Matching papers

Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation

667 upvotes · 15 OCT 2026 · Bingxi Hou, Guochao Jiang, Guofeng Quan et al.

This paper develops a method to transfer policy shifts from a reinforcement learning (RL) teacher to a student model, using an intermediate teacher checkpoint to compare early and late policy shifts and gate the endpoint shift based on their agreement. Practitioners in RL and reinforcement learning may care about this paper because it proposes a more effective way to transfer policy shifts, which is crucial for improving student performance in various domains.

Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability

153 upvotes · 6 OCT 2026 · Bingxi Hou, Guochao Jiang, Guofeng Quan et al.

This paper investigates how to improve the effectiveness of On-Policy Distillation, a technique that trains a student model on its own generated data with teacher feedback. By prioritizing supervision reliability over alignment coverage, the researchers found that using a smaller subset of shared vocabulary at aligned positions can lead to comparable accuracy to using the full shared vocabulary.

When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation

96 upvotes · 17 SEP 2026 · Yuxiao Yang, Tianrun Yu, Shangzhe Li et al.

This paper investigates why some AI models can generate excessively long responses and proposes a solution to mitigate this issue by aligning the models' termination tokens. Practitioners might care about this problem because it can lead to wasted generation budgets and inefficient AI applications.