RuntimeError: mat1 and mat2 must have the same dtype, but got Float and BFloat16

[louis-fashion-ai]

I use V100, so I only use float32 model, but I get this error, can you solve this problem?

[helsenIgnace]

I had the same error, turned out an 'with torch.cuda.amp.autocast(dtype=torch.float32)'. Be sure to check whether you changed all the torch.cuda.amp.autocast() to use torch.float32 instead of bfloat16.