alex-petrenko/sample-factory

Health v2 · maintenance only

Adoption not part of the score

1017 stars · 153 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Sample Factory is a high-throughput Python reinforcement learning library implementing synchronous and asynchronous policy gradient algorithms (PPO/APPO). It supports single- and multi-agent training, self-play, Population-Based Training, and integrates with environments like MuJoCo, Atari, ViZDoom, Isaac Gym, and DeepMind Lab.

Use cases

• train RL agents with PPO at high throughput
• train multi-agent RL policies with self-play
• train agents in ViZDoom or Isaac Gym environments
• run population-based training for RL experiments
• benchmark RL algorithm throughput on CPU or GPU
• upload trained RL models to the HuggingFace Hub

When to choose

• you need maximum training throughput for policy gradient RL
• you want synchronous or asynchronous PPO with multi-GPU support
• you need multi-agent training, self-play, or PBT
• you want a library designed to be imported with custom environments as first-class citizens

When to avoid

• you need Windows support
• you need off-policy algorithms like SAC or DQN
• you want a simple high-level API over many algorithm families rather than a focused PPO implementation

Facets

library · maturity active

reinforcement-learning machine-learning gpu-computing benchmarking reinforcement-learning machine-learning artificial-intelligence robotics python ppo appo policy-gradient multi-agent self-play population-based-training vizdoom isaacgym mujoco atari huggingface game-development linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="alex-petrenko/sample-factory")