[CI/Build] Fixed the xla nightly issue report in #12451 #12453
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can do one of these:
🚀 |
|
cc @lsy323 |
DarkLight1337
changed the title
DarkLight1337
changed the title
github-actions
bot
added
the
ready
|
Cancelled the CI as TPU tests are failing, please fix them. |
Head branch was pushed to by a user without write access
|
Thanks @DarkLight1337 - fixed the issue - the tests should pass now |
…the requirements.txt is fixed to solve the vLLM on TPU issue Signed-off-by: Hossein Sarshar [email protected]
| torch_xla[pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.7.0.dev20250124-cp39-cp39-linux_x86_64.whl ; python_version == "3.9" | ||
| torch_xla[pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.7.0.dev20250124-cp310-cp310-linux_x86_64.whl ; python_version == "3.10" | ||
| torch_xla[pallas] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.7.0.dev20250124-cp311-cp311-linux_x86_64.whl ; python_version == "3.11" | ||
| torch==2.6.0.dev20241216+cpu |
There was a problem hiding this comment.
torchxla nightly and torch nightly may not be compatible, due to c++ symbol issue. Please see the previous comment for detail.
…llm-project#12453) Signed-off-by: Isotr0py <[email protected]>
|
@hosseinsarshar and @lsy323 the TPU runner is consistently failing in CI now due to the nightly version being removed from the index. I assume these are cleared over time. https://buildkite.com/vllm/ci/builds/13854#01952464-d826-4e96-8085-806bfa5483c7/42-148 Can you please look at fixing? Thanks |
…llm-project#12453) Signed-off-by: Linkun Chen <[email protected]>
The PR fixes the issue reported in bug #12451
FIX #12451
Context:
Apparently, there is an issue with the nightly XLA 2.7 release that is incompatible with the upstream torch 2.7, causing vLLM to break on TPUs.
I also reported this issue on the XLA repo (pytorch/xla#8626)
Resolution:
Matching the working combination - xla 2.7 dev20250124 with torch 2.6 dev20241216