# PaddlePaddle/PARL

A high-performance distributed training framework for Reinforcement Learning

Repository: https://github.com/PaddlePaddle/PARL
Canonical: https://ross.abutalabs.com/products/parl
Homepage: https://parl.readthedocs.io/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: reinforcement-learning, large-scale, parallelization
Last push: 2025-09-13T06:29:18+00:00

## Health v2 (maintenance only)
Score: 41/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 41, release rhythm 8, longevity 100
- inputs: {"age_days": 3052, "days_push": 354, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3453, forks 812 (observed 2026-08-28T04:08:05.362703+00:00)

## What it is
PARL is a flexible, high-performance reinforcement learning framework built on PaddlePaddle, providing Model/Algorithm/Agent abstractions and reproducible implementations of influential RL algorithms. Its xparl component enables large-scale distributed training across thousands of CPUs and multiple GPUs with a simple decorator-based API.

## Use cases
- train reinforcement learning agents at scale
- parallelize RL training across thousands of CPUs
- reproduce classic RL algorithms like DQN and PPO
- build custom RL algorithms by inheriting base classes
- apply GPU-accelerated RL to autonomous driving simulation
- run distributed multi-GPU reinforcement learning experiments

## When to choose
- you need scalable, distributed RL training with minimal code changes
- you want reproducible implementations of standard RL algorithms
- you are already using the PaddlePaddle ecosystem
- you need to parallelize environment data collection across many CPU workers

## When to avoid
- you prefer PyTorch- or TensorFlow-based RL libraries like Stable-Baselines3 or RLlib
- you only need small-scale single-machine RL without parallelization
- you need supervised or self-supervised learning rather than RL

## Facets
- artifact type: framework
- maturity: active
- function: reinforcement-learning, machine-learning, llm-training
- domain: reinforcement-learning, machine-learning, deep-learning
- platform: python, cross-platform
- tags: distributed-training, parallelization, paddlepaddle, rl-algorithms, xparl, linux, gpu

## Member repositories
- PaddlePaddle/PARL (main) score 41

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:05.362703+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 2026-08-29T18:37:32.414709+00:00, confidence not recorded.
  - readme: https://github.com/PaddlePaddle/PARL (fetched 2026-08-28T04:08:05.362703+00:00, sha 8efdc16eaee5)
  - registry_pypi: https://pypi.org/pypi/parl/json (fetched 2026-08-29T09:31:48.690094+00:00, sha 850f0500218a)
- Data as of 2026-08-30T08:39:29.467469+00:00.