Hugging Face's logo

Datasets:
aps
/
super_glue

Tasks:
Text Classification
Token Classification
Question Answering
Sub-tasks:
natural-language-inference
word-sense-disambiguation
coreference-resolution
Languages:
English
Size:
10K<n<100K
ArXiv:
Tags:
superglue
NLU
natural language understanding
License:
Dataset card Files Community
14

Why are the labels -1 in test data

#7
by srija616 - opened
Discussion
srija616

Hi! I am new to NLP and I am finetuning copa on a bert model already fine-tuned on social_i_qa dataset. I don't understand why labels for test data of copa are all -1. Shouldn't they be 0 or 1 (depending on choice1 and choice2)?
Please tell me what am I missing here?

Davidg707

@srija616 only the train and validation sets have "the answers" (the labels). There are no public values for the test set. The way to find out your performance on the test set is to upload it to the SuperGLUE site.

albertvillanova

Thanks for your accurate answer, @Davidg707 .

I'm closing this discussion.

albertvillanova changed discussion status to closed

Sign up or log in to comment