Skip to content

When turning a pre-trained model into an LSC model, how to decide when to start stage 2 training? #2

Description

@Darkbblue

Hello! This is a great study. I find HunyuanDiT especially stable compared to DiTs without LSCs, but this model is a bit outdated, so I want to add LSCs to other DiTs myself.
I managed to add some LSC adapters to a pre-trained DiT. Interestingly, I notice that although the generated image is largely corrupted, I can still observe some patterns of the overall object contour.
Anyway, my major concern now is how to decide when to start the stage 2 training, a.k.a. when you train the entire model. Should I wait until the generated image quality is almost restored before going into stage 2?

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

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions