Have you changed training hyper-parameters after replacing info-nce with your rince?

[CoinCheung]

You trained with batch size of 4096 and learing rate is (4096/256)*0.3=4.8 which is different from mocov3. Did I understand the code correctly?

I did not find training args of mocov2-resnet50 in the paper(maybe I was too careless to find them), would you please share your training hyper-parameters?