[Benchmarks] Make detokenization optional in benchmark scripts

[JArnoldAMD]

The benchmark scripts currently include detokenization as part of the performance measurement. In most cases the detokenization time is minimal, but for some models and script parameters it can be significant. Since detokenization isn't always needed (e.g. it may be handled by a separate server), it is helpful to have the option of whether or not to include detokenization in the measurement.

Since benchmark scripts for some other frameworks (notably TensorRT-LLM) do not include detokenization time, I have opted to skip detokenization by default. This default could be changed if desired.

[github-actions[bot]]

👋 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.

🚀

[mergify[bot]]

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @JArnoldAMD.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[JArnoldAMD]

This is a reasonable idea. I do think we should maintain the current behavior by default, which is to include detokenization in these benchmarks

I'll update to change the default to do include detokenization, with a flag to disable it.

[njhill]

Now also supported in V1: https://github.com/vllm-project/vllm/pull/14224