Proposal to add Qwen3-TTS support

[ShahVandit]

Model description

Model: Qwen3-TTS
Repository: https://github.com/QwenLM/Qwen3-TTS
Paper/Blog: https://huggingface.co/papers/2601.15621
License: Apache 2.0

Qwen3-TTS is currently available as a standalone package (qwen-tts) but is not integrated into transformers. The model uses PreTrainedModel and GenerationMixin, making it compatible with the transformers architecture.

Would the maintainers be open to this addition? I'm happy to submit a PR if this aligns with transformers' roadmap?

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

No response

[Rocketknight1]

cc @eustlb @ebezzam

[farzanehnakhaee70]

I am also intersted on adding this model to transformer library. vllm-omni currently has been added this model to his online-inference server. However, as it is not supported by the transformers, it is not useable yet.

[ShahVandit]

Hi! I'm implementing the Qwen3-TTS integration into transformers. Configuration, modeling, and processing utilities are complete. Currently working on tests to ensure everything works correctly.

[ebezzam]

@ShahVandit that's great! please open a PR when it's ready. Here are a few resources to help you with the Transformers integration:

• Every model integration should use modular to define (at least) the modeling, and sometimes configuration/processing: https://huggingface.co/docs/transformers/en/modular_transformers
• A few recent audio model additions so you can get an idea of the files involved and coding style: GLM-ASR and VibeVoice (ongoing but also a TTS model, and conventions are often changing so best to see latest approaches)

[pjgao]

Any progress for this model?

[ShahVandit]

Qwen3-TTS implementation is complete and ready for review.
#44517

[ShahVandit]

Hi @ebezzam , just checking in on the PR when you get a chance. Happy to make any changes based on your feedback!

[ebezzam]

Hi @ShahVandit thanks for the message! I haven't forgotten, there's a few ongoing model additions so I kept postponing my review 🙃 I'll try to drop some pointers tomorrow or early next week. Thanks for your contribution 🙏