Closed shuzi closed 5 years ago
@shuzi see https://github.com/szagoruyko/wide-residual-networks/tree/fp16 branch and set data_type=torch.CudaHalfTensor th train.lua
to train in float16
@szagoruyko
just curious, when you use FP16 on P100 card, the optimization still converges?
@shuzi I don't have a P100 card.
Models in pytorch hub were trained in FP16 https://pytorch.org/hub/pytorch_vision_wide_resnet/
Hi , would you please point me to the part of code that can switch between FP32 mode and FP16 mode ?