luchris429/purejaxrl

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1099 stars · 87 forks observed · 2026-08-28

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

PureJaxRL provides end-to-end reinforcement learning training pipelines implemented entirely in JAX, including environments, enabling massive speedups via JIT compilation and vectorization. It offers CleanRL-style single-file implementations like PPO that can run thousands of parallel agent seeds on a single GPU.

Use cases

• train PPO agents on GPU with JAX
• run thousands of RL seeds in parallel for hyperparameter tuning
• speed up reinforcement learning training 1000x over PyTorch
• do meta-evolution and meta-RL research
• learn single-file RL algorithm implementations
• benchmark RL algorithms on MinAtar and Cartpole

When to choose

• you need extremely fast RL training with many parallel seeds on GPUs
• you want to jit/vmap/pmap entire RL training pipelines including environments
• you're doing RL research like meta-evolution or rapid hyperparameter sweeps
• you prefer CleanRL-style readable single-file implementations

When to avoid

• you need a modular, importable RL library for production applications
• you require CPU-only training or non-JAX ecosystems like PyTorch
• you need a wide algorithm coverage beyond the provided PPO variants
• you want stable APIs and long-term maintenance guarantees

Facets

library · maturity active

reinforcement-learning machine-learning gpu-computing benchmarking reinforcement-learning deep-learning machine-learning python cloud jax ppo single-file-implementations meta-reinforcement-learning jit-compilation research-code research gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="luchris429/purejaxrl")