Skip to content

Conversation

@zpcore
Copy link
Member

@zpcore zpcore commented Jul 31, 2024

Resolves #7730.

Now device are automatically detected when we import torch_xla:

$ python
Python 3.10.13 | packaged by conda-forge | (main, Dec 23 2023, 15:36:39) [GCC 12.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch_xla
WARNING:root:libtpu.so and TPU device found. Setting PJRT_DEVICE=TPU.
>>> torch_xla.real_devices()
['TPU:0', 'TPU:1', 'TPU:2', 'TPU:3']
>>> torch_xla.device()
device(type='xla', index=0)

@zpcore zpcore marked this pull request as ready for review July 31, 2024 20:05
@zpcore zpcore requested review from JackCaoG and will-cromar July 31, 2024 20:05
@zpcore
Copy link
Member Author

zpcore commented Jul 31, 2024

RFC: Is there a plan to hard code PJRT_DEVICE in the python script to replace the PJRT_DEVICE environment variable? @miladm

Copy link
Collaborator

@will-cromar will-cromar left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the clean up!

@JackCaoG do you see any problem selecting $PJRT_DEVICE when we import torch_xla?

@zpcore zpcore added usability Bugs/features related to improving the usability of PyTorch/XLA tpuci labels Jul 31, 2024
@JackCaoG
Copy link
Collaborator

Thanks for the clean up!

@JackCaoG do you see any problem selecting $PJRT_DEVICE when we import torch_xla?

seems fine to me as long as we don't init the runtime yet.

@zpcore zpcore requested a review from will-cromar August 1, 2024 18:01
@zpcore zpcore merged commit 2ba4e05 into master Aug 1, 2024
@zpcore zpcore deleted the piz/pjrt branch September 3, 2024 17:29
byi8220 added a commit to byi8220/pytorch-lightning that referenced this pull request Nov 23, 2024
Fixes Lightning-AI#20419

`torch_xla.runtime.using_pjrt()` is removed in pytorch/xla#7787

This PR replaces references to that function with a check to [`device_type()`](https://github.com/pytorch/xla/blob/master/torch_xla/runtime.py#L83) to recreate the behavior of that function, minus the manual initialization
lantiga pushed a commit to Lightning-AI/pytorch-lightning that referenced this pull request Nov 25, 2024
…py for `torch-xla>=2.5` (#20442)

* Replace `using_pjrt()` xla runtime `device_type()` check with in xla.py

Fixes #20419

`torch_xla.runtime.using_pjrt()` is removed in pytorch/xla#7787

This PR replaces references to that function with a check to [`device_type()`](https://github.com/pytorch/xla/blob/master/torch_xla/runtime.py#L83) to recreate the behavior of that function, minus the manual initialization

* Added tests/refactored for version compat

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* precommit

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

usability Bugs/features related to improving the usability of PyTorch/XLA

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Improve device auto-detection

4 participants