Support PyTorch main commits & vLLM PRs

[BoyuanFeng]

Currently the CI supports PyTorch stable (2.7.1) and pulls latest docker image from vLLM.

Feature requests:

• Support PyTorch main commit. So we can track regression due to PyTorch side.
• Support vLLM commits from forker repo or PR. vLLM github repo requires contributors to fork the repo and create a PR. It does not allow creating a branch in vLLM github repo.

[fadara01]

@huydhn @BoyuanFeng do you guys know of any workflow to trigger the vllm benchmarks from vLLM PRs?

[fadara01]

Currently the CI supports PyTorch stable (2.7.1) and pulls latest docker image from vLLM.

@BoyuanFeng could you please explain what do you mean by Currently the CI supports PyTorch stable (2.7.1) ?
Wouldn't the relevant PyTorch version be whatever it's set in vLLM to be? Am I missing something?

[huydhn]

If you have write access to this repo, you can dispatch vLLM benchmark manually via https://github.com/pytorch/pytorch-integration-testing/actions/workflows/vllm-benchmark.yml. We keep it this way because there isn't enough capacity on OSS to run benchmark on all commits / PRs by default