google-deepmind/acme
Acme
Acme is a reinforcement learning framework and execution engine designed for developing and benchmarking learning algorithms. It provides a library of modular components and reference implementations used to construct agents and establish performance baselines.
The system enables the scaling of agent architectures from single-stream execution to large distributed environments. This allows for the transition from initial prototyping to distributed execution for training and evaluation.
The framework covers reinforcement learning development and agent architecture prototyping, providing the building blocks necessary for benchmarking new models against standard reference agents.
Features
- Distributed RL Scaling - Acts as an execution engine for scaling reinforcement learning training and rollout generation across distributed GPU nodes.
- Reinforcement Learning - Offers a comprehensive framework for training agents through trial and error using modular components.
- Distributed - Scales reinforcement learning workflows from single-stream processes to large distributed systems.
- RL Agent Implementation Frameworks - Provides the structural support and modular building blocks necessary for implementing RL agents.
- Agentic Reinforcement Learning - Provides a specialized framework for developing and training agents using reinforcement learning.
- Reinforcement Learning Agents - Provides a library of reference implementations for constructing reinforcement learning agents.
- Distributed Execution Scaling - Enables scaling agent architectures from single-stream execution to large distributed environments.
- RL Algorithm Benchmarking Toolkits - Implements toolkits for benchmarking new reinforcement learning algorithms against standard reference agents.
- Agent Architectures - Supports the design and iteration of structural patterns for reinforcement learning agents.
- Reinforcement Learning - Library of reinforcement learning components and agents.
