Skip to content

The Replacement of Open-source JudgeModel/Evaluator #19

@bittersweet1999

Description

@bittersweet1999

Hi WildBench Team 👋,

We have updated a series of Fine-Tuned Judge Models:
opencompass/CompassJudger-1-32B-Instruct
opencompass/CompassJudger-1-14B-Instruct
opencompass/CompassJudger-1-7B-Instruct
opencompass/CompassJudger-1-1.5B-Instruct

In relevant experiments, we tested the reliability of these models as Judgers. Among them, CJ-1-14B and 32B achieved a Judge correlation of over 95% with GPT4o-0806 on the WildBench dataset, which can be considered as a low-cost alternative to GPT4o. For more details, please refer to our paper. https://arxiv.org/pdf/2410.16256

Thank you!

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