[doc] add warning about comparing hf and vllm outputs

[youkaichao]

A lesson learned from #1069 (comment) .

In our ci, we always use do_sample=False for huggingface models, so it is not a problem.

[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]

Let's merge this first. Later we can implement #10758

[youkaichao]

oh i don't know #10758 . Is that what we want to implement? that's kind of bc-breaking.

[DarkLight1337]

I would argue that #10758 is the "intended" behavior. To avoid breaking changes, we can log a warning for a while before finally switching over to using generation_config.json as the default.

[youkaichao]

generation_config.json is aligned with huggingface generate function, if we take default values from the file, the implementation will be coupled.

[DarkLight1337]

Let's move the discussion to that issue so we can gather thoughts from the OP as well.