Faquad-nli needs to be updated

#18
by nicholasKluge - opened

Hi Eduardo!

As of datasets==4.0, loading scripts and trust_remote_code are no longer supported.

This breaks things, like the lm-evaluation-harness-pt, which people who work with Portuguese LLMs need for running evals.

I opened a request/issue on Ruan's dataset repository (issue), with a script on how to convert the current faquad-nli into something datasets supports.

But as of now, running the harness on faquad_nli won't work. I made a working copy of the dataset just so I can keep working (nicholasKluge/faquad-nli-parquet). As soon as Ruan updates ruanchaves/faquad-nli, I'll delete my copy.

If anyone else needs to run the harness while the original dataset is not converted, feel free to use my copy.

eduagarcia pinned discussion

Sign up or log in to comment