16-bit (torch.HalfTensor) support possible to save VRAM?

crowsonkb / style-transfer-pytorch

Neural style transfer in PyTorch.

MIT License

479 stars 115 forks source link

16-bit (torch.HalfTensor) support possible to save VRAM? #21

Open arpitest opened 2 years ago

arpitest commented 2 years ago

Is it possible to do the processing using 16-bit float type, using model.half() and torch.HalfTensor data type? it could lower the VRAM requirements for high resolution images.