What does this PR do ?

Add Qwen3‑Omni training and example scaffolding with smoke conversion/inference, plus a new L0 CI launcher for the Qwen3‑Omni conversion test.

Changelog

• Add Qwen3‑Omni example scripts and README (conversion, inference, local training) with repo‑local workspace defaults.
• Add Qwen3‑Omni smoke conversion and inference flow (HF → Megatron → HF export).
• Add L0 CI entrypoint for Qwen3‑Omni conversion tests with repo‑root coverage paths.
• Add Qwen3‑Omni unit and functional tests (recipe registry, conversion, smoke validation).

GitHub Actions CI

See the CI sectionin the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• This is PR2; PR1 is currently under review: [model] Add initial Qwen3-Omni support in Megatron-Bridge #2831
• This PR is independent of PR1 and focuses on example/test scaffolding only. It does not introduce changes to the core model implementation.
• The training recipe has been validated on 4 nodes × 8 GPUs with initial performance tuning.
• Follow-up work: performance/efficiency optimizations will be handled in a separate PR.
• Related to # (issue)

Summary by CodeRabbit

New Features

• Qwen3-Omni Model Integration: Added full support for Qwen3-Omni multimodal model with text, image, video, and audio capabilities.
• Checkpoint Conversion Tools: New utilities to convert checkpoints between Hugging Face and Megatron formats.
• Training Examples: Added example scripts and recipes for fine-tuning the model with multimodal data.
• Inference Workflows: New example scripts demonstrating single-rank and distributed inference scenarios.

Documentation

• Added comprehensive model documentation including supported features and known limitations.
• Added training plan documentation outlining staged implementation roadmap.
• Added README files documenting example usage and workspace setup.