Early-Signal Detector

method · first seen 4 Oct 2026

on-policy distillation

Also: OPD, on-policy knowledge distillation, Open Publication Distribution System, multi-objective on-policy distillation

Distilling a student model from teacher signals computed on the student's own generated trajectories.

On-policy distillation trains a student model on trajectories it generates itself, with a teacher scoring or supplying token-level distributions over those student-produced sequences, rather than imitating fixed teacher outputs. The sources show it being treated as a standard post-training step for reasoning LLMs and then extended in many directions: sparse teacher interfaces for long traces (ResOPD), multi-teacher merging (Janus-MOPD), block diffusion and few-step image diffusion students, speculative decoding drafts (a Qwen repo), agent memory and ReAct behavior, multimodal perception, and backdoor transfer from teacher to student. Activity is almost entirely arXiv papers from many independent academic and industry groups. A few hits, such as an OPDS e-book server, are acronym collisions.

If it keeps spreading, on-policy distillation could become a default way to compress and merge frontier models into smaller ones. That would make its failure modes practical concerns for anyone shipping distilled models, including teacher errors on student-induced states, multi-domain interference and inherited backdoors.

Analyst note (emerging), 17 Nov 2026.

Mentions per week

04719 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 12 on 16 Nov 2026 · 78 mentions in 28 days · 77 independent groups · papers → code · first seen 6 weeks ago · acceleration +0.08

Earliest sources

Latest sources

Mentioned together with

knowledge distillationgroup relative policy optimizationreinforcement learning with verifiable rewardsself-distillationspeculative decodingreverse KL divergencevision-language-action modelflow matchingdiffusion language modelmulti-teacher distillation

In the ledger