[Frontend] Add LLM.reward specific to reward models

[noooop]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

After #21470, LLM.encode pooling_task defaults to using "encode", meaning that reward models actually use LLM.encode. However, most libraries' encode corresponds to an embedding model, which can lead to potential errors.

• Add LLM.reward specified to reward models
• The pooling_task for LLM.encode defaults to None, and pooling_task needs to be specified

Test Plan

Test Result

(Optional) Documentation Update

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

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 either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Code Review

This pull request aims to improve the API for reward models by adding a dedicated LLM.reward method and making the pooling_task in LLM.encode explicit. The changes are in the right direction, but the implementation of the new reward method seems incorrect as it uses pooling_task="encode" and lacks necessary validation. I've left a critical comment with suggestions to align it with other task-specific methods like embed and classify. Additionally, I've suggested improving an error message for better user guidance. Addressing these points will significantly improve the correctness and consistency of the new API.

[noooop]

@DarkLight1337

Quick review

[DarkLight1337]

I think we should still keep encode in VllmRunner just like for LLM class

[noooop]

I think we should still keep encode in VllmRunner just like for LLM class

Perhaps we need to figure out whether it requires embed or reward.

[DarkLight1337]

This model uses pooler interface for image segmentation. So it's not really embed/reward. The more generic encode (which we can rename later) is a better match.

[noooop]

I'll wait for buildkite/ci/pr/language-models-test-extended-pooling to finish and check for any other errors

[DarkLight1337]

Let's wait for #18321 to be merged first, then update this PR

[noooop]

buildkite/ci/pr/language-models-test-extended-pooling

The failed test can be fixed by #21747