Skip to content

[Trainer] refactor/consolidate trainer implementations #624

@casteryh

Description

@casteryh

Context

There are code duplications in our implementations of the trainer.
For example, there is torchforge/src/forge/actors/trainer/titan.py and there is a standalone sft trainer implementation in apps/sft/main.py.

To do

  • refactor our trainer implementation in torchforge/src/forge/actors/trainer/titan.py to support, for example, SFT
  • use the new TitanTrainer in the SFT app

Metadata

Metadata

Labels

best practicesThings we should be doing but aren'tbetter engineeringTasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions