Acme is a Python RL research framework with modular building blocks and agents. It supports installation via pip and offers examples, tutorials, and documentation for distributed RL research.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Acme is a library of reinforcement learning components and agents designed as building blocks for research. It provides reference implementations and flexible bases for running agents at multiple scales, including single-stream and distributed configurations.
How it works
Acme exposes agents and components intended for easy instantiation and experimentation. The README outlines working code examples in the examples subdirectory and notebooks for quickstart and tutorials. It emphasizes that agents can be run with either JAX or TensorFlow dependencies, depending on the agent.
Getting started
Installation guidance is provided via a step-by-step workflow:
- Create a virtual environment and install core tooling:
python3 -m venv acme
source acme/bin/activate
pip install --upgrade pip setuptools wheel
- Install core library with optional dependencies for agents (JAX or TF):
pip install dm-acme[jax,tf]
- Install example environments:
pip install dm-acme[envs]
- For bleeding-edge from GitHub, install from the repository with extra options:
pip install .[jax,tf,testing,envs]
Recent releases
Latest releases include:
- 0.4.0 v0.4.0 (2022-02-10)
- 0.3.0 v0.3.0 (2022-01-08)
- 0.2.4 0.2.4 (2021-12-09)
- 0.2.3 0.2.3 (2021-12-03)
- 0.2.2 0.2.2 (2021-08-27)
Notes under Highlights mention dependency updates for TensorFlow and Reverb and a new SVG-0 agent.
Traction
Stars: 4036 (GitHub stars). Open issues: 101. Language: Python. License: Apache-2.0.
Behind the repo
Not included in the provided facts.
Caveats
License: Apache-2.0. Created: 2020-05-01. Last push: 2026-08-01. Open issues: 101. No further caveats listed in the provided materials.