Skip to content

How much performance improvement does RL bring? #3

Open
Open
How much performance improvement does RL bring?#3

Description

@richyuan10
richyuan10
opened on Sep 18, 2025

could you share how much improvement the RL training adds over the SFT model?

Activity

  1. linqingwang commented on Sep 18, 2025

    @linqingwang
    linqingwang
    on Sep 18, 2025
    Collaborator

    Actually, the primary function of SFT is to provide the correct prior knowledge, ensuring a standardized output:{<think....</think\n reprompt...}; the main performance improvements come from GRPO.

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