Fix text-to-speech pipeline crash when generation config contains None values

[jiqing-feng]

Description

The text-to-speech pipeline crashes when loading models whose generation_config contains fields set to None (e.g. use_cache=None). This is because TextToAudioPipeline.__init__ blindly copies all generation config values into the model config via config.update(gen_config.to_dict()), which triggers huggingface_hub's strict dataclass field validation when a None value is assigned to a typed field (e.g. bool).

The fix filters out None values before updating the config.

Reproduction

from transformers import pipeline

synthesiser = pipeline("text-to-speech", "sesame/csm-1b", device="cpu")

huggingface_hub.errors.StrictDataclassFieldValidationError: Validation error for field 'use_cache':
    TypeError: Field 'use_cache' expected bool, got NoneType (value: None)

Fix

- config.update(gen_config.to_dict())
+ config.update({k: v for k, v in gen_config.to_dict().items() if v is not None})

[jiqing-feng]

Hi @ydshieh . Would you please review this PR? Thanks!

[jiqing-feng]

Hi @Rocketknight1 . Would you please review this PR? Thanks!

[Rocketknight1]

Yeah, makes sense to me! None values probably do not want to override any existing config settings anyway.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.