validation set accuracy

[saar-eliad]

Hi,
after switching PyTorch version + using an updated model from huggingFace after they fixed a small bug,
I noticed a small 0.5% drop in our validation set accuracy.
I want to compare it to your numbers - however, I did not find your validation set results.
Are they available somewhere?

Can you share them for T5-3B super-glue: WIC, BoolQ, MultiRC, RTE?
(the results I'm getting are very close to what you published on the test set).