I am using a pretrained model from transformers which expects input type to be torch.HalfTensor
. However, I have input of type torch.FloatTensor
. How to convert?
torch.Tensor.half()
should do:
torch.rand(3,3).half()
# tensor([[0.1456, 0.0580, 0.9277],
# [0.3318, 0.1979, 0.6670],
# [0.5391, 0.2012, 0.2445]], # dtype=torch.float16)