Skip to content

Cross benchmark contamination #9

@ilyasch2

Description

@ilyasch2

Hi,
Thank you very much for this work.
I am trying to assess the contamination between different benchmarks. For example, between mmlu training set and mmlu-pro validation set (and vice versa). However, different benchmarks might have different format i.e different keys. And we can set only a global text_keys arg. Thus we run into the assertion error
Is there any way to set different text keys/key currently?

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