Add flag to enable float32 computation for normalization (norm + affine)#2536

rwightman merged 2 commits intomainhuggingface/pytorch-image-models:mainfrom

fp32_normhuggingface/pytorch-image-models:fp32_normCopy head branch name to clipboard