All Tools
P
AgentsFreeOpen Source
PUREJAXRL
End-to-end JAX implementations of deep RL algorithms
Apache-2.0
ABOUT
Typical RL stacks bounce between Python env steps and a separate learner, so GPU time is wasted on host round-trips. PureJaxRL keeps PPO and the environment in one JAX program that compiles end to end, which makes large sweeps practical on accelerators.
INTEGRATION GUIDE
1. Train PPO end-to-end in JAX on Gymnax, Brax, or MinAtar environments
2. Benchmark how fast a compiled RL loop runs versus a PyTorch baseline
3. Prototype a new JAX agent by editing a single-file PureJaxRL implementation
4. Run GPU or TPU RL experiments without a CPU environment bottleneck
TAGS
pythonjaxreinforcement-learningppogpuresearch