Text-to-audio pipeline failed in 5.17.0

[zhangwei217245]

System Info

The text-to-audio pipeline failed when I run this pipeline using the suno/bark-small model as well as the facebook/musicgen-stereo-small

For these two models, the to() function is not expecting a 'dtype' parameter.

I didn't test other models, but pretty much I believe this is an issue!

File "/app/venvs/flexserv/lib/python3.14/site-packages/transformers/pipelines/text_to_audio.py", line 297, in __call__
    return super().__call__(text_inputs, **forward_params)
           ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/venvs/flexserv/lib/python3.14/site-packages/transformers/pipelines/base.py", line 1299, in __call__
    return self.run_single(inputs, preprocess_params, forward_params, postprocess_params)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/venvs/flexserv/lib/python3.14/site-packages/transformers/pipelines/base.py", line 1302, in run_single
    model_inputs = self.preprocess(inputs, **preprocess_params)
  File "/app/venvs/flexserv/lib/python3.14/site-packages/transformers/pipelines/text_to_audio.py", line 197, in preprocess
    output = output.to(dtype=self.model.dtype)

I am attaching the links to the related PR here:

#40546

640a08a#diff-df2480d65db99054fa0a1b6d7275165f2b91beac00a435a19a545eb2c269f561R196-R197

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Just install transformers library 5.17.0 and write a python program to call text-to-audio pipeline to run suno/bark-small model and you will see the error.

Expected behavior

The pipeline should run cleanly instead of a failure.

[vinitsonawane45]

Hi @zhangwei217245, I’d like to investigate and work on this issue.

I can reproduce the failure with the affected text-to-audio models and will trace the regression back to the TextToAudioPipeline preprocessing change introduced in #40546. I’ll also add regression coverage for the affected models/path and verify that the fix does not break the other text-to-audio models.

If there are no objections, I’ll work on a PR for this.