Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A curated collection of papers and resources on On-Policy Distillation for Large Language Models.
| Date | Stars |
|---|---|
| 2026-08-05 | 494 |
| 2026-08-14 | 510 |
| 2026-08-18 | 513 |
| 2026-08-19 | 518 |
| 2026-08-20 | 519 |
| 2026-08-21 | 521 |
| 2026-08-22 | 522 |
| 2026-08-23 | 521 |
| 2026-08-25 | 522 |
| 2026-08-27 | 523 |
| 2026-08-28 | 524 |
| 2026-08-30 | 525 |
| 2026-09-02 | 527 |
| 2026-09-03 | 529 |
| 2026-09-05 | 533 |
| 2026-09-06 | 534 |
| 2026-09-07 | 535 |
| 2026-09-09 | 537 |
| 2026-09-10 | 538 |
| 2026-09-11 | 539 |
| 2026-09-12 | 540 |
| 2026-09-14 | 541 |
| 2026-09-16 | 543 |
| 2026-09-17 | 546 |
| 2026-09-18 | 547 |
| 2026-09-19 | 548 |
| 2026-09-20 | 551 |
| 2026-09-21 | 552 |
| 2026-09-22 | 555 |
| 2026-09-23 | 558 |
| 2026-09-27 | 559 |
| 2026-09-29 | 561 |
| 2026-09-30 | 562 |
| 2026-10-02 | 563 |
| 2026-10-03 | 565 |
| 2026-10-04 | 566 |
| 2026-10-05 | 567 |
| 2026-10-06 | 569 |
| 2026-10-07 | 569 |
| 2026-10-08 | 570 |
| 2026-10-09 | 571 |
| 2026-10-10 | 573 |
| 2026-10-11 | 575 |
| 2026-10-12 | 578 |
| 2026-10-13 | 580 |
| 2026-10-18 | 581 |
| 2026-10-19 | 582 |
| 2026-10-20 | 584 |
| 2026-10-23 | 587 |
| 2026-10-24 | 587 |
| 2026-10-27 | 588 |
| 2026-10-29 | 589 |
| 2026-10-31 | 590 |
| 2026-11-01 | 592 |
| 2026-11-04 | 592 |
Today
— stars today
This week
+4 stars this week
This month
+25 stars this month
Momentum
4.0
growth rate 0.68%/day
<a id="readme-top"></a>
<p align="center">
<a href="https://nick7nlp.github.io/OPDHub/">
<picture>
<source media="(prefers-color-scheme: dark) and (max-width: 600px)" srcset="assets/reading-room-mobile-dark.svg">
<source media="(max-width: 600px)" srcset="assets/reading-room-mobile-light.svg">
<source media="(prefers-color-scheme: dark)" srcset="assets/reading-room-dark.svg">
<img src="assets/reading-room-light.svg" width="1200" alt="Awesome LLM On-Policy Distillation — learning from teachers on the student’s own trajectories.">
</picture>
</a>
</p>
<p align="center">
<a href="https://arxiv.org/abs/2604.00626"><strong>Read the survey</strong></a> · <a href="https://nick7nlp.github.io/OPDHub/"><strong>Explore OPDHub ↗</strong></a> · <a href="#browse-the-literature">Browse papers</a> · <a href="#citation">Cite</a>
</p>
<p align="center"><sub>362 entries · 134 with code · Updated 2026‑10‑13</sub></p>
A reading companion to **[A Survey of On-Policy Distillation for Large Language Models](https://arxiv.org/abs/2604.00626)**.
<a id="a-few-good-starting-points"></a>
## Start reading
<table>
<thead><tr>
<th align="left" width="440">Follow a question</th>
<th align="left" width="440">Start with</th>
</tr></thead>
<tbody>
<tr>
<td><strong>How does the student learn?</strong><br><sub>Sampling, divergence, and the policy update.</sub></td>
<td><a href="https://arxiv.org/abs/2306.13649">GKD</a> · <a href="https://arxiv.org/abs/2306.08543">MiniLLM</a></td>
</tr>
<tr>
<td><strong>Where does supervision come from?</strong><br><sub>Solutions and feedback as teacher context.</sub></td>
<td><a href="https://arxiv.org/abs/2601.18734">OPSD</a> · <a href="https://arxiv.org/abs/2601.20802">SDPO</a></td>
</tr>
<tr>
<td><strong>What transfers, and what does not?</strong><br><sub>Reasoning compatibility, skills, and knowledge.</sub></td>
<td><a href="https://arxiv.org/abs/2604.13016">Rethinking OPD</a> · <a href="https://arxiv.org/abs/2610.09639">Skills vs. Knowledge</a></td>
</tr>
<tr>
<td><strong>How do multiple agents learn?</strong><br><sub>Joint trajectories and merging distilled experts.</sub></td>
<td><a href="https://arxiv.org/abs/2609.34234">MAS-OPD</a> · <a href="https://arxiv.org/abs/2609.34225">USA</a></td>
</tr>
</tbody>
</table>
**[Follow the full reading paths →](resources/reading-order.md)**
<a id="the-collection"></a>
<a id="find-your-question"></a>
## Browse the literature
Expand a topic to read the paper summaries, or **[search and filter on OPDHub ↗](https://nick7nlp.github.io/OPDHub/)**. Entries are ordered newest first within each topic.
<a id="papers-4"></a>
<a id="objectives-and-update-rules"></a>
<details><summary><strong>Objectives and update rules</strong> · 64</summary>
*What does the student update optimize?*
- **[Residual Advantage: Student-Relative Teacher Guidance for RL with Verifiable Rewards](https://arxiv.org/abs/2610.11519)**
<sub>2026-10-08 · Teacher-mediated RL</sub>
RA derives a local score update with a conditional squared-distance matching identity, then combines response-centered residual guidance with outcome RL. Its controls and Co-RA study distinguish teacher usefulness from standalone solver accuracy; the full clipped update is not exact L2 minimization.
- **[PIVOT: Perplexity-Informed KD-to-RL Transition Scheduling for Vertical-Domain Few-Shot Distillation](https://arxiv.org/abs/2610.11167)**
<sub>2026-10-08 · Hybrid OPD</sub>
PIVOT routes student-generated examples between teacher-distribution matching and GRPO using teacher-evaluated perplexity; on two few-shot intent-classification tasks, its step-matched recipe outperformed continued OPD and a global OPD-to-GRPO switch, while routing ablations favored sending lower-perplexity examples to GRPO earlier.
- **[Towards Unbiased On-Policy Distillation for Block Diffusion Language Models](https://arxiv.org/abs/2610.05373)** Excerpt of 187,622 characters
Read on GitHub
196
43
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:aa33f5bf80cd647e, topic:large-language-models, topic:llm
matched fp:aa33f5bf80cd647e, topic:knowledge-distillation
matched fp:aa33f5bf80cd647e, topic:rlhf
matched fp:aa33f5bf80cd647e, topic:awesome-list