# BytedTsinghua-SIA/DAPO

An Open-source RL System from ByteDance Seed and Tsinghua AIR

Repository: https://github.com/BytedTsinghua-SIA/DAPO
Canonical: https://ross.abutalabs.com/products/dapo
Language: Python
License Family: other
Last push: 2025-05-11T09:09:41+00:00

## Health v2 (maintenance only)
Score: 29/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 21, release rhythm 35, longevity 38
- inputs: {"age_days": 534, "days_push": 479, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1861, forks 85 (observed 2026-08-28T04:05:45.392189+00:00)

## What it is
DAPO is an open-source reinforcement learning system for large-scale LLM training, released by ByteDance Seed and Tsinghua AIR. It implements the Decoupled Clip and Dynamic Sampling Policy Optimization algorithm on top of the verl framework, with code, datasets, and model weights achieving 50% on AIME 2024.

## Use cases
- train llm with reinforcement learning
- reproduce dapo rl training on qwen models
- run rlhf-style policy optimization for math reasoning
- evaluate a 32b model on aime 2024
- research scalable rl for large language models

## When to choose
- you want to reproduce or extend state-of-the-art LLM RL training
- you need the DAPO algorithm, dataset, and checkpoints together
- you already use the verl framework and want proven RL recipes

## When to avoid
- you need a production-ready training platform with support guarantees
- you lack multi-GPU infrastructure for large-scale RL
- you need a permissively licensed codebase, since no license is specified

## Facets
- artifact type: library
- maturity: active
- function: llm-training, reinforcement-learning, machine-learning, benchmarking
- domain: large-language-models, reinforcement-learning, deep-learning, machine-learning
- platform: python
- tags: rlhf, policy-optimization, verl, math-reasoning, aime, research-code, gpu, linux, docker

## Member repositories
- BytedTsinghua-SIA/DAPO (main) score 29

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:45.392189+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:15:48.840027+00:00, confidence not recorded.
  - readme: https://github.com/BytedTsinghua-SIA/DAPO (fetched 2026-08-28T04:05:45.392189+00:00, sha c819b142273c)
- Data as of 2026-08-30T08:39:29.467469+00:00.