pufferlib
Version-aware guidance for PufferLib reinforcement-learning environments
by k-dense-aiFree
Description
pufferlib is an agent skill that gives version-aware guidance for PufferLib, a library for reinforcement-learning environments. It covers vectorisation, policies, PuffeRL training, evaluation and safe review of checkpoints. The skill starts with a point that matters a lot: choose the version before choosing an API. The current native 5.0 source is for C and CUDA environments and a native trainer, with CPU evaluation only. The latest PyPI release, pufferlib 3.0.0 from 23 June 2025, is source distribution only and covers Python, Gymnasium and PettingZoo adaptation with the Torch PuffeRL trainer. A pinned 4.0 snapshot is kept as history, with a C Ocean interface and an optional Torch fallback. For 5.0 the skill points to a native reference file. Its command line works as ./puffer train after an environment is built, not as puffer train with an environment name. There is no 5.0 Python emulation or vector API, and no slow fallback flag. The review was dated 1 October 2026. Use it when you are building or training a reinforcement-learning agent with PufferLib and need the correct API for your version.…
Highlights
- Version-aware guidance for PufferLib
- Native 5.0 C and CUDA environments
- PuffeRL training and evaluation
- Safe checkpoint review
- Separate paths for 3.0 and 4.0
Ratings & Reviews
No ratings yet
Be the first to rate pufferlib.
People also visit
pytorch-patterns
PyTorch deep learning patterns and best practices for building robust
pufferlib FAQ
What is PufferLib?
A library for reinforcement-learning environments, vectorisation, policies and PuffeRL training.
Why does the version matter?
The skill says the APIs differ across versions, so the version must be chosen before the API. Version 5.0 has no Python emulation or vector API.
How do I run training in 5.0?
Build the environment first, then use ./puffer train. The skill says not to use puffer train with an environment name.
Which version is on PyPI?
pufferlib 3.0.0, published 23 June 2025, distributed as a source package only.
How do I install it?
Run npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill pufferlib.
Is it checked for security?
Its security audits show passes from Gen Agent Trust Hub, Socket and Snyk.
What are the best pufferlib alternatives?
Popular alternatives to pufferlib include pytorch-patterns, aeon and cirq. See all pufferlib alternatives