TorchBeast: A PyTorch Platform for Distributed RL

TL;DR

TorchBeast is an open-source PyTorch platform that simplifies scalable reinforcement learning research by implementing IMPALA with both single-machine and multi-machine versions, maintaining high performance and ease of use.

Contribution

It introduces a flexible, easy-to-use RL platform in PyTorch with both pure Python and high-performance multi-machine implementations, facilitating scalable RL research.

Findings

Performs on par with IMPALA on Atari benchmarks

Provides both Python-only and multi-machine high-performance versions

Enables scalable RL research with minimal programming complexity

Abstract

TorchBeast is a platform for reinforcement learning (RL) research in PyTorch. It implements a version of the popular IMPALA algorithm for fast, asynchronous, parallel training of RL agents. Additionally, TorchBeast has simplicity as an explicit design goal: We provide both a pure-Python implementation ("MonoBeast") as well as a multi-machine high-performance version ("PolyBeast"). In the latter, parts of the implementation are written in C++, but all parts pertaining to machine learning are kept in simple Python using PyTorch, with the environments provided using the OpenAI Gym interface. This enables researchers to conduct scalable RL research using TorchBeast without any programming knowledge beyond Python and PyTorch. In this paper, we describe the TorchBeast design principles and implementation and demonstrate that it performs on-par with IMPALA on Atari. TorchBeast is released as an open-source package under the Apache 2.0 license and is available at \url{https://github.com/facebookresearch/torchbeast}.