please use release with torch / tpu versions about xla

[dlibenzi]

As I mentioned, the PyTorch/TPU project is under heavy development, and we have switched the architecture from using the PyTorch/JIT, to using a new XTen (lazy tensor) approach.
In order for you to fetch a version working with TF 1.13, you will have to fetch the PyTorch and PyTorch/XLA commit IDs before Dec 19th 2018, which is a LONG time ago.
And you will be fetching a version using the old JIT approach, which has limited support (if you are planning to train MNIST or resnets, that is OK - anything more, you will likely find "unsupported operator" errors).

[dlibenzi]

Hi!
Thanks for the interest in the PyTorch/TPU project. The project is under heavy development, and unfortunately, as we move forward, we discover new features that we need on the TPU VM side (the TF service).
We expect the project to be stabilizing in the new couple of months, and we are also speeding up the TF release process, so that the gap between cutting edge TF, and the version inside the TPU VM, will be smaller.

[ildoonet]

@dlibenzi Thanks for the reply! So do you have any recommendation of versions to use current xla? (eg. xla commit id, torch commit id, tpu pod tf version, ...)

[ildoonet]

@dlibenzi Thanks for the hard work you put in but current repository is somewhat misguiding many people. It is not installable with the current version of readme. Please provide commit ids(pytorch, tensorflow, xla) and tpu pod version that we can test.

[ildoonet]

Thanks I will look forward to your improved works!