Question regarding the batch size

[won5830]

Hi thank you for your wonderful contribution.
I'm currently trying to retrain the model using your data with mixed precision.
However, I have a question about your batch size regarding CellViT initiated with SAM-B, H, and L.
What was your batch size for the training ? was it all 1 ?
I found that you used 16 for in your supplementary data but I found that GPU run out of its memory even with 80 GB memory.
Is there any more things to pay attention other than mixed precision for training such large model ?

[FabianHoerst]

On which dataset do you try to train the network? We used 16 as the batch-size for PanNuke.