All Tools
S
AgentsFreeOpen Source
SAMPLE FACTORY
High-throughput synchronous and asynchronous reinforcement learning
MIT
ABOUT
Standard RL trainers bottleneck on sampling, so experiments on fast simulators still wait on a single process. Sample Factory runs synchronous and asynchronous PPO at high throughput across many workers, letting teams collect millions of steps per second without rewriting the algorithm.
INSTALL
pip install sample-factoryINTEGRATION GUIDE
1. Train high-throughput PPO on VizDoom or Isaac Gym with many parallel workers
2. Compare synchronous versus asynchronous sample collection on the same policy
3. Scale a custom Gymnasium environment to millions of steps without a custom distributed trainer
4. Reproduce sample-efficient versus sample-hungry RL baselines with a shared fast collector
TAGS
pythonpytorchreinforcement-learningppodistributedhigh-throughput