search

alex-petrenko / sample-factory

High throughput synchronous and asynchronous reinforcement learning
https://samplefactory.dev
MIT License
811 stars 109 forks source link

Added Multi GPU #254

Closed Mikahil closed 1 year ago

Mikahil commented 1 year ago

Added cfg entry "gpu_per_policy" which defaults to 1 and creates batchers, samplers, and learners to match the value. When its greater than one, the learners will use DistributedDataParallel to synchronize the policy's parameters.

alex-petrenko commented 1 year ago

@Mikahil if not too much trouble, consider running make format next time before committing. This will make sure pre-commit checks pass and you can actually run the tests on GH actions.

https://www.samplefactory.dev/12-community/contribution/#how-to-create-a-pull-request