Skip to content

How to fine-tune DiffCoder? #2

@osyvokon

Description

@osyvokon

Hi ByteDance! First of all, great job on Stable-DiffCoder! 🚀

I'm trying to understand the recommended way to fine-tune the model for a custom use case. As far as I understand, traditional auto-regressive frameworks like trl doesn't match the pretraining objective and won't work or, worse, will silently degrade performance.

Is there an official or recommended fine-tuning pipeline?

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