[Model] Deprecate the score task (this will not affect users).

[noooop]

Purpose

Cross-encoder (aka reranker) models are a subset of classification models that accept two prompts as input and output num_labels equal to 1.

In the engine, we will no longer differentiate between classify and score. So we can deprecate the score task.

This is part of deprecate support for multi-task.

Test Plan

Test Result

---

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.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

The pull request successfully deprecates the 'score' task, replacing it with 'classify' where appropriate and removing related logic and comments. The changes are consistent across tests, model configuration, entrypoints, and model executor components. The refactoring of activation functions and renaming of variables like pooled_data to embeddings or logits enhances code clarity and maintainability. No critical or high-severity issues were found in the changes.

[mergify]

Hi @noooop, the pre-commit checks have failed. Please run:

uv pip install pre-commit>=4.5.1
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy failing?

mypy is run differently in CI. If the failure is related to this check, please use the following command to run it locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10

[mergify]

Documentation preview: https://vllm--37537.org.readthedocs.build/en/37537/

[robertgshaw2-redhat]

@noooop - this broke CI
@DarkLight1337 - please ensure we run relevant tests

[noooop]

@noooop - this broke CI
@DarkLight1337 - please ensure we run relevant tests

test below passed successfully when landing.

https://buildkite.com/vllm/ci/builds/56995/steps/canvas?jid=019d058c-c6a8-44d4-90b5-adbb3c3bfe7f&tab=output

let's rerun this test.

• [Model] Deprecate the score task (this will not affect users).  #37537

https://buildkite.com/vllm/ci/builds/57220/steps/canvas?sid=019d09f2-b0ab-4b55-abae-5ffcd500c440&tab=output

https://buildkite.com/vllm/ci/builds/57231/steps/canvas?sid=019d0a4a-8d65-4473-94ce-9944c14d21be&tab=output

• [ROCm][CI] Fix accuracy for llama-nemotron-vl pooling tests #37613

https://buildkite.com/vllm/ci/builds/57288/steps/canvas?sid=019d0be7-c922-4bac-9e36-1dae8d24be1a&tab=output

https://buildkite.com/vllm/ci/builds/57302/steps/canvas?sid=019d0c41-4cef-4034-a5ea-8a40b9a97275&tab=output

All these tests passed.

I also cannot reproduce this error locally. That's too strange.

[noooop]

[noooop]

[noooop]

Perhaps it was caused by #37303, and it should be fixable by #37775.

cc @robertgshaw2-redhat @AndreasKaratzas

[AndreasKaratzas]

@noooop i can open it, i just want also #37774 to be merged so that amd ci is evaluated as well. Can we merge 37774 first?