Skip to content

Modify neural methods's __call__ to do just one step #512

Description

@michalk8

Currently, the __call__ method takes in various dataloaders. In order to have more flexible way of training (e.g. for sharding), the __call__ should do just one step of the algorithm.
In a separate PR later, we will then define a generic trainer.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions