Skip to content

Can I apply incremental learning here? What suggestions do you have? #967

@probiewh

Description

@probiewh

I have trained a Pi0.5 LoRA model on 200 episodes for 40,000 steps, and it works well in most cases. However, during inference, there are still a few error cases. I have collected new correction data for these specific failure cases.
Can I perform incremental LoRA fine-tuning based on my already trained 200-episode model?
If yes, could you please give me some advice on the number of training steps for the new small dataset, relative to my original 40,000 steps?
If I re-train the model from scratch on the full dataset every time I collect new data, it will take an extremely long time. So I really want to use incremental learning efficiently.
Thank you so much for your help!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions