Spaces:

Yeyito

/

llm_contamination_detector

like 25

Running

I'm fairly certain my Winogrande implementation in "/detect-pretrain-code-contamination/src/eval.py -> process_winogrande()" is incorrect. I've been uncertain of it since the beginning and having almost every model score 0.0-0.01 on this test doesn't give me the confidence to publish these scores publicly.

I'm unsure of how to make a correct implementation of this, would love it if someone could open a PR.