[build][misc] allow to use recent numpy

[zhouyuan]

This patch removed the limitation on numpy < 2.0.0
outlines already did a workaround for running with recent numpy
dottxt-ai/outlines#1265

Fixes #6570

[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.

🚀

[DarkLight1337]

I think we also need to update the compiled requirements-test.txt?

[DarkLight1337]

I think you need to pip compile with -P numpy to keep the numpy version in the output txt

[zhouyuan]

I think you need to pip compile with -P numpy to keep the numpy version in the output txt

hi @DarkLight1337
it seems genai-perf has a requirement for old numpy, so pip compile will stick to 1.26.4
https://github.com/triton-inference-server/perf_analyzer/blob/main/genai-perf/pyproject.toml#L51
The genai-perf is only used in nightly benchmarks, but it's marked as one dependency in test. So I think we should try to bump the numpy in genai-perf package firstly?

-yuan

[DarkLight1337]

cc @robertgshaw2-redhat

[mgoin]

@debermudez @jikunshang could you help us understand if genai-perf can update to recent numpy?

[debermudez]

@mgoin this actually came up today in a different thread for our team. I think another team mate is already working on it.

[zhouyuan]

Yes, here's the feedback from genai-perf dev
triton-inference-server/perf_analyzer#272 (comment)

[dyastremsky]

Done, thanks for raising this! triton-inference-server/perf_analyzer#275

[mergify]

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

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

[rgommers]

genai-perf 0.0.11 was released with numpy 2 support five days ago, so this PR should be unblocked. Could you please rebase @zhouyuan? It'd be great to get this in.

[zhouyuan]

genai-perf 0.0.11 was released with numpy 2 support five days ago, so this PR should be unblocked. Could you please rebase @zhouyuan? It'd be great to get this in.

@rgommers Yes, genai-perf is fine now, there seems to have legacy code in tritonclient which still have this restriction. I have made a quick patch here:

triton-inference-server/client#826

[mergify]

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

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

[mgoin]

Resolved by another PR

[zhouyuan]

Hi @mgoin it looks like numpy==1.26.4 is still a requirement here. Not sure if I was looking at the wrong place.
https://github.com/vllm-project/vllm/blob/main/requirements/test.txt#L318

Thanks, -yuan

[mgoin]

That is for testing, which has many more deps than the package itself have. There is no restriction on the common numpy version anymore https://github.com/vllm-project/vllm/blob/main/requirements/common.txt