Skip to content

Models that were trained from a previous finetune #54

Description

@johny-b

Suppose I trained GPT-4.1 on something for 3 epochs and then trained the resulting model again (i.e. using the previous one as starting point) for 2 epochs.

I have a model trained for 5 epochs that started from GPT-4.1. Yet

manager.get_model_list(base_model="gpt-4.1", epochs=5)

won't show it, because in the database the base model is the model we used when starting second-stage training, i.e. the model after the first stage.

Now, if both these models are in the same project, this could be solved - just find the real base model using the previous stage model.

But maybe it will be confusing? What if someone trains on the second dataset for 5 epochs and gives the same suffix? Then, these models will be very different but look the same.
(Perhaps having the same suffix for trains-from-base and trains-from-previous-ft should be forbidden)

TBD what to do here.

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

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions