--- title: On-Policy Distillation | daily.dev description: On-policy distillation combines the strengths of reinforcement learning (on-policy sampling) and supervised fine-tuning (dense reward signal) for post-training... canonical: https://daily.dev/posts/on-policy-distillation-mlnm2ztfi twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: On-Policy Distillation | daily.dev og:description: On-policy distillation combines the strengths of reinforcement learning (on-policy sampling) and supervised fine-tuning (dense reward signal) for post-training... og:url: https://daily.dev/posts/on-policy-distillation-mlnm2ztfi og:image: https://api.daily.dev/og/posts/mlNM2ZtfI.png og:image:alt: On-Policy Distillation og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # On-Policy Distillation **[Thinking Machines Lab](https://daily.dev/sources/thinkingmachines)** · 28 min read · 0 upvotes · 0 comments ## Summary On-policy distillation combines the strengths of reinforcement learning (on-policy sampling) and supervised fine-tuning (dense reward signal) for post-training LLMs. By sampling trajectories from a student model and using a larger teacher model to grade each token via reverse KL divergence, it achieves 9-30x compute savings over off-policy distillation and roughly 10x over RL. Experiments show it reaches 70% on AIME'24 math benchmarks in ~150 steps from a 60% SFT checkpoint, and effectively recovers instruction-following behavior lost during domain-specific fine-tuning. The approach is also promising for continual learning, as it avoids the policy drift that plagues SFT when training on a model's own outputs. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Papers Explained 595: Unsupervised On-Policy Self Distillation](https://daily.dev/posts/papers-explained-595-unsupervised-on-policy-self-distillation-zvvvb9e9v) · Medium · 1 upvotes · 0 comments - [Researchers propose a self-distillation fix for ‘catastrophic forgetting’ in LLMs](https://daily.dev/posts/researchers-propose-a-self-distillation-fix-for-catastrophic-forgetting-in-llms-cfeweepzh) · InfoWorld · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#reinforcement-learning](https://daily.dev/tags/reinforcement-learning) [View this post on daily.dev](https://daily.dev/posts/on-policy-distillation-mlnm2ztfi) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"On-Policy Distillation","url":"https://daily.dev/posts/on-policy-distillation-mlnm2ztfi","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/on-policy-distillation-mlnm2ztfi"},"datePublished":"2026-08-02T07:05:07.582Z","dateModified":"2026-08-02T07:32:30.959Z","description":"On-policy distillation combines the strengths of reinforcement learning (on-policy sampling) and supervised fine-tuning (dense reward signal) for post-training...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/10ce0df6ef23115d661b95aae5bdd721?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/10ce0df6ef23115d661b95aae5bdd721?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Thinking Machines Lab","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Thinking Machines Lab","logo":"https://media.daily.dev/image/upload/s--1Y_UQWQ9--/f_auto,q_auto/v1785654300/logos/thinkingmachines","url":"https://daily.dev/sources/thinkingmachines"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/on-policy-distillation-mlnm2ztfi","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,reinforcement-learning","timeRequired":"PT28M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Thinking Machines Lab","item":"https://daily.dev/sources/thinkingmachines"},{"@type":"ListItem","position":3,"name":"On-Policy Distillation"}]} ```