The Problem

Teams building large‑language‑model pipelines need a single, up‑to‑date reference that lists the literature, reports, and open‑source implementations for on‑policy distillation (OPD). Without such a curated source, engineers waste time hunting scattered papers and code bases, risking duplicated effort and missed best‑practice guidance.

What This Does

The repository is a curated list of OPD resources. The root contains only markdown‑style documentation:

  • README.md – the primary entry point; it renders the badge header, a short description of OPD, and a hierarchical table of contents that links to sections such as Surveys, Core OPD Papers, and Frameworks and Implementations.
  • CITATION.cff – supplies a citation file for downstream users who want to reference the list in academic work.
  • CONTRIBUTING.md – outlines how external contributors can add entries, update categories, or fix formatting.
  • LICENSE – an MIT‑style license granting unrestricted reuse.
  • .claude/skills/add-paper/SKILL.md – a Claude‑specific skill definition used by the author’s internal tooling for adding new paper entries (not part of the public API).

No executable code, scripts, or data files are present; the repo’s purpose is purely informational.

How It Is Wired

Because the repository contains only static markdown and metadata files, there is no runtime wiring to describe. The only “entry point” is the README.md, which is rendered by GitHub (or any markdown viewer) and serves as the navigation hub. No functions, imports, or external calls exist, so a call‑graph cannot be generated. Consequently, the repository has no side effects, no file‑system or network interactions, and no internal dependencies beyond the markdown renderer provided by the hosting platform.

How To Use It

# Clone the repository
git clone https://github.com/moses-y/awesome-on-policy-distillation.git
cd awesome-on-policy-distillation
  1. Open README.md in a markdown viewer or on GitHub.
  2. Use the table of contents to jump to the desired section (e.g., Core OPD Papers → Foundations).
  3. Follow the listed URLs to access the original papers, technical reports, or implementation repositories.
  4. To contribute a new entry, edit README.md following the format described in CONTRIBUTING.md, then submit a pull request.

No build step, environment variables, or runtime command are required.

Real‑World Use

A research engineer at a startup can clone the repo, locate the latest OPD survey, and immediately pull the cited GitHub implementations (e.g., Qwen3 or Nemotron‑Cascade 2) for benchmarking. The engineer can then add a new internal benchmark result to the list by editing README.md per the contribution guide, keeping the team’s knowledge base synchronized with the broader community.

Code Health & Issues

  • Medium – No test files detected – repository-wide (static analysis).
  • Medium – No CI/CD pipeline detected – repository-wide (no .github/ workflows, no Makefile, no Dockerfile).
  • Low – Documentation limited to markdown – only three doc files (README.md, CONTRIBUTING.md, CITATION.cff).
  • Low – No lockfile – dependency management not applicable (no code).

All findings are derived from deterministic static analysis; no additional issues were inferred.

The Bottom Line

awesome-on-policy-distillation is a well‑structured, MIT‑licensed markdown collection that centralizes OPD literature and tooling references. It offers immediate value for researchers and engineers needing a single source of truth, but it provides no executable component, and the project lacks automated testing or CI. Ideal for teams that want a curated bibliography rather than a codebase to integrate directly.