PufferAI/PufferLib

Health v2 · maintenance only

Adoption not part of the score

6306 stars · 548 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

PufferLib is a fast, open-source reinforcement learning library that trains tiny, super-human models in seconds, achieving 1M+ environment steps per second. It bundles its own learning algorithms, hyperparameter tuning, and high-performance C/CUDA environment simulation methods.

Use cases

• train reinforcement learning agents on game environments at millions of steps per second
• build high-performance custom RL environments in C or CUDA
• run hyperparameter tuning for RL experiments
• train tiny super-human models quickly for research
• run RL demos client-side in the browser
• benchmark and compare RL training methods

When to choose

• you need maximum RL training throughput on game-like environments
• you want a batteries-included library with algorithms, tuning, and fast environments
• you are doing RL research and want reproducible, fast baselines

When to avoid

• you need a general-purpose deep learning framework rather than RL
• your environments are slow Python-only and cannot be vectorized
• you need a mature ecosystem of third-party integrations comparable to Gym/Stable-Baselines3

Facets

library · maturity active

reinforcement-learning machine-learning simulation gpu-computing reinforcement-learning machine-learning gpu-computing python cpp c cross-platform rl-training environment-vectorization high-performance game-environments pufferlib game-development gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="PufferAI/PufferLib")