cfivek

I might be able to help you, i was dealing with some memory issues as well but sorted it out with different batch sizes / mixed precision/swap memory settings. I...

I didnt get the email, but did you try running nvidia-smi in the cmd window to see what's eating up the memory? If the GPU is only 12gb I would...

> It says the same as for ir50 and ir_101 > > How does incresing classnum increases memory consumption? Increasing number of classes increases the output layer and needs more...

> > > It says the same as for ir50 and ir_101 > > > How does incresing classnum increases memory consumption? > > > > Increasing number of classes...

This would be very nice to have

I have gone back and tested each login for myself. The only ones that work for me, are Google, Paypal, and Hotmail/Live. Wondering if this is the same for others?...

The way around this memory leak is to run this line encoding, _ = model(bgr_tensor_input) using with torch.no_grad() Does doing this affect the inference results in any way?