Skip to content

Storing metadata for fine-tuned models #85

Closed

Description

@eyaler

I am fine tuning various models which have different styles of prompts. In inference I need to style the prompts adequately for each model. It would be great if it was possible to save a metadata or config json when creating the model, and retrieve it with the model. Otherwise I would need to either: (1) manage my own external database which is cumbersome when using google colab (2) hack the model suffix to encode my config params (let's not do that), or (3) reverse engineer my training jsonl to deduce the model's specific prompt styling, which is difficult. Any other options?

Activity

  1. hallacy commented on May 18, 2022

    @hallacy
    Contributor

    That sounds like a reasonably summary of the situation to me. I'll pass this request on to the fine tuning team

  2. hallacy commented on Mar 7, 2023

    @hallacy
    Contributor

    We don't plan to build out this feature, so closing the issue.

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

    enhancementNew feature or requestwontfixThis will not be worked on

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions