[NPU][Bugfix] Align GPU side and recover qwen3-tts

[gcanlin]

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

Purpose

FIX #1508.

As a workaround, we have to add a hardware-specific code in modeling file. I will remove it quickly in the next release.

This PR also updates the docs to prepare the coming release v0.16.0.

Test Plan

 vllm serve Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice --omni --port 8091 --stage-configs-path /root/vllm-workspace/vllm-omni/vllm_omni/platforms/npu/stage_configs/qwen3_tts.yaml

curl -X POST http://localhost:8091/v1/audio/speech     -H "Content-Type: application/json"     -d '{
        "input": "Hello, how are you?",
        "voice": "vivian",
        "language": "English"
    }' --output output.wav

Test Result

• Qwen3-Omni
• Qwen3-TTS
• Wan2.2 with hsdp

---

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 Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8be2f6e3b7

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Avoid shadowing attn_utils in build_attn_metadata

Importing attn_utils inside the NPU-only branch makes attn_utils a local variable for the whole function. When current_omni_platform.is_npu() is false (e.g., GPU/CPU execution), the else branch calls attn_utils.build_attn_metadata(...) before that local is assigned, which raises UnboundLocalError and breaks the Qwen3-TTS code predictor path on non-NPU backends.

Useful? React with 👍 / 👎.

[gcanlin]

Good catch. Fixed!

[gcanlin]

cc @hsliuustc0106 @Gaohan123 PTAL, thanks!

[lishunyang12]

Left a couple of minor comments. The TTS/Omni unification in the model runner looks solid overall — the getattr-based dispatch and the cudagraph guard make sense.