[CI][Bugfix] Improve cosine similarity calculation by incorporating length harmony adjustment in text comparison

[yenuo26]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

To address the insensitivity issue in similarity calculation exposed by #2953 and #2812 when dealing with repeated text patterns in test case.

Problem Analysis:
The n-gram counting cosine similarity treats multiple occurrences of "ABC" in "ABCABCABC" as part of the vector, still considering it "directionally similar" to a single "ABC". This results in an inflated similarity score where length differences are not properly reflected, making it difficult to flag as anomalous.

Solution:
In cosine_similarity_text, introduce a length-imbalance-sensitive factor (e.g., F1-style harmonic mean of lengths) multiplied by the original cosine similarity. This approach keeps the behavior unchanged for equal-length texts while suppressing similarity scores in such repetitive cases.

Test Plan

1. run cosine_similarity_text in local

cosine_similarity_text("Beijing, the capital of China, is a city where millennia of history seamlessly blend with breathtaking modernity. As one of the world's oldest continuously inhabited cities, it serves as the nation's political, cultural, and educational heart.Beijing, the capital of China, is a city where millennia of history seamlessly blend with breathtaking modernity. As one of the world's oldest continuously inhabited cities, it serves as the nation's political, cultural, and educational heart.", "Beijing, the capital of China, is a city where millennia of history seamlessly blend with breathtaking modernity. As one of the world's oldest continuously inhabited cities, it serves as the nation's political, cultural, and educational heart.")

2.run test case in CI

Test Result

1.local

• before

• after

2.ci

---

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. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[yenuo26]

@amy-why-3459 PTAL

[amy-why-3459]

Thank you so much for your quick fix. To prevent CI from continuing to fail, we need to merge the bug fix before merging this change.