[WIP][CI] Add CI for Ming Omni

[edwingao28]

Motivation

Closes #295. This PR builds the multi-stage Ming-flash-omni-2.0 CI suite mirroring the qwen3-omni structure and covers the following metrics: TTFA, RTF, WER, plus MMMU and MMSU accuracy

Modification

Workflow DAG

  docs ──► stage-1-thinker ──► stage-2-tts
                            ├─► stage-3-mmmu
                            └─► stage-4-mmsu

stage-1-thinker — text generation

Pure text path, thinker only. tests/test_model/test_ming_omni_thinker_length.py exercises Ming thinker across short / medium / long prompts and asserts response shape.

stage-2-tts — text-to-speech

RTF performance + non-stream TTFA baseline diagnostic + WER accuracy. Single job with two tests sharing one server (test_tts_non_streaming_perf + test_tts_wer).

tests/test_model/test_ming_omni_tts_ci.py uses voice_clone=False

stage-3-mmmu — image recognition

Image+text → text via thinker only. tests/test_model/test_ming_omni_mmmu_ci.py runs mmmu-ci-50 with warmup=2

stage-4-mmsu — audio-in understanding

Text+audio → text via MMSU benchmark with modalities="text+audio". This covers the #295 ASR requirement as a superset tests/test_model/test_ming_omni_mmsu_ci.py runs thinker-only (talker OFF).

Accuracy Test

Benchmark & Profiling

benchmark & ci threshold in progress

Checklist

• Format your code according with pre-commit.
• Add unit tests.
• Update documentation / docstrings / example tutorials as needed.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.