Skip to content

[TPU] update torch_xla pin #19231

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jun 6, 2025
Merged

[TPU] update torch_xla pin #19231

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
acfaec4
[TPU] update torch_xla pin
yaochengji Jun 5, 2025
97425d6
fix LIBTPU_INIT_ARGS
yaochengji Jun 6, 2025
6e31886
fix pre-submit
yaochengji Jun 6, 2025
751e463
fix pre-submit
yaochengji Jun 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 5 additions & 5 deletions requirements/tpu.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,9 @@ setuptools==78.1.0
--find-links https://storage.googleapis.com/libtpu-releases/index.html
--find-links https://storage.googleapis.com/jax-releases/jax_nightly_releases.html
--find-links https://storage.googleapis.com/jax-releases/jaxlib_nightly_releases.html
torch==2.8.0.dev20250529
torchvision==0.22.0.dev20250529
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250529-cp39-cp39-linux_x86_64.whl ; python_version == "3.9"
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250529-cp310-cp310-linux_x86_64.whl ; python_version == "3.10"
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250529-cp311-cp311-linux_x86_64.whl ; python_version == "3.11"
torch==2.8.0.dev20250605
torchvision==0.23.0.dev20250605
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250605-cp39-cp39-linux_x86_64.whl ; python_version == "3.9"
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250605-cp310-cp310-linux_x86_64.whl ; python_version == "3.10"
torch_xla[tpu, pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250605-cp311-cp311-linux_x86_64.whl ; python_version == "3.11"

2 changes: 1 addition & 1 deletion tests/tpu/test_moe_pallas.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
# The Pallas GMM kernel requires num_tokens * topk to be a multiple of 16
@pytest.mark.parametrize("m", [8, 16, 64, 2048])
@pytest.mark.parametrize("n", [128, 1024, 2048])
@pytest.mark.parametrize("k", [128, 512, 1024])
@pytest.mark.parametrize("k", [128, 511, 1024])
Copy link
Contributor

Choose a reason for hiding this comment

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

medium

Consider adding a test case where k is chosen such that num_tokens * topk is a multiple of 16, given the constraint mentioned in line 27. This would provide more confidence in the kernel's correctness under the required condition.

@pytest.mark.parametrize("e", NUM_EXPERTS)
@pytest.mark.parametrize("topk", TOP_KS)
@pytest.mark.parametrize("ep_size", EP_SIZE)
Expand Down
3 changes: 2 additions & 1 deletion vllm/v1/worker/tpu_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,7 +100,8 @@ def init_device(self):
# `xla_tpu_force_1d_allreduce_at_chunk_count` is a temporary solution to
# fix this. It will be removed after the bug in XLA compiler is fixed.
os.environ["LIBTPU_INIT_ARGS"] = (
"--xla_tpu_force_1d_allreduce_at_chunk_count=1")
os.environ.get("LIBTPU_INIT_ARGS", "") +
" --xla_tpu_force_1d_allreduce_at_chunk_count=1")
Copy link
Collaborator

Choose a reason for hiding this comment

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

nit: We can add a comment saying the additional libtpu arg is needed due to pytorch/xla#9084

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

I think it's fine. Because here we're not adding any specific libtpu arg, but inherit all the args if any.

Copy link
Collaborator

Choose a reason for hiding this comment

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

I see, makes sense, thanks!

torch.set_grad_enabled(False)
torch.set_default_dtype(self.model_config.dtype)

Expand Down