[CI][Installation] Avoid uploading CUDA 11.8 wheel

[cermeng]

I found the latest wheel from https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl is built from CUDA 11.8 which is inconsistent with documentation https://docs.vllm.ai/en/latest/getting_started/installation.html#install-the-latest-code

I think this is due to https://github.com/vllm-project/vllm/blob/main/.buildkite/upload-wheels.sh#L36-L38. Since both the CUDA 11.8 and 12.1 release pipelines will use this script to upload the wheel to S3. The pipeline that completes later will overwrite the previous upload. Then it may be random whether CUDA 11.8 or CUDA 12.1 is used, depending on which version builds faster.

Fix it by not uploading the wheel of CUDA 11.8

cc @simon-mo

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

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:

• Add ready label to the PR
• Enable auto-merge.

🚀

[mgoin]

Good catch!

[simon-mo]

Please don't merge yet.

[simon-mo]

The cu11 wheel should have the right suffix in the wheel name. Something must changed...

[simon-mo]

testing my changes here https://buildkite.com/vllm/release/builds/1929

[simon-mo]

seems to work.

aws s3 cp artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/32a354491655819eba03f329fe8977a0530c44e2/
--
  | upload: artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/32a354491655819eba03f329fe8977a0530c44e2/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
  | [2024-11-21T21:02:58Z] + aws s3 cp artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/nightly/
  | upload: artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
  | [2024-11-21T21:03:00Z] + aws s3 cp artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/1.0.0.dev/
  | upload: artifacts/dist/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/1.0.0.dev/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl



[2024-11-21T20:44:42Z] + aws s3 cp artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/32a354491655819eba03f329fe8977a0530c44e2/
--
  | upload: artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/32a354491655819eba03f329fe8977a0530c44e2/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl
  | [2024-11-21T20:44:44Z] + aws s3 cp artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/nightly/
  | upload: artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/nightly/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl
  | [2024-11-21T20:44:46Z] + aws s3 cp artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl s3://vllm-wheels/1.0.0.dev+cu118/
  | upload: artifacts/dist/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl to s3://vllm-wheels/1.0.0.dev+cu118/vllm-1.0.0.dev+cu118-cp38-abi3-manylinux1_x86_64.whl

https://buildkite.com/vllm/release/builds/1929#0193506f-2078-4ddc-954d-ec3a9df8134c/120-3878

[cermeng]

The cu11 wheel should have the right suffix in the wheel name. Something must changed...

Just out of curiosity, is there a public link to access the cu11.8 wheel?

[sleepwalker2017]

@simon-mo How can we get wheel for CUDA 11.8 now?