PyTorch 1.3.0 has been released

Highlights include:

  1. Mobile support
  2. Named tensors
  3. Quantization
  4. Type Promotion
  5. New TorchScript/JIT features and improvements
  6. TPU support

It will be great to include some of these features in fastai


Yes it would be in particular amazing if we were able to deploy models to mobile as easily as in web apps!

named tensors, type promotion and captum integration look very interesting

I trained resnet18 from scratch with their Tpu support in like 10 min. I think Tpu will be the way forward for colab notebooks.

1 Like