Fix Audio.cast_storage for CSV-loaded large_string columns

[pdTetteh]

Summary

This PR fixes Audio.cast_storage() so that audio path columns loaded as large_string can be cast to Audio() correctly.

Motivation

Dataset.from_dict({"audio": [...]}) produces a string column and cast_column("audio", Audio()) works, but load_dataset("csv", ...) produces a large_string column, which currently fails during the cast to the Audio storage type.

Changes made

• Added handling for pa.large_string() in Audio.cast_storage()
• Cast large string storage to pa.string() before building the Audio struct storage
• Added a regression test covering the CSV-loaded audio path case

Testing

• Reproduced the issue locally
• Ran the targeted regression test for the CSV-loaded large_string case
• Ran tests/features/test_audio.py in the local environment

Closes #7970

[lhoestq]

Thanks for the investigation and the fix ! This makes me think that maybe load_dataset("csv") should return "string" types instead of "large_string" to be consistent with load_dataset("json") as well as the Dataset.from_xxx methods

[lhoestq]

The proposed fix looks good to me to enable Audio on large_string types :)

Could you also add the same fix for Video and Pdf ?

[lhoestq]

you can use the same code as in Image.cast_storage

Suggested change

	if pa.types.is_string(storage.type) or pa.types.is_large_string(storage.type):
	if pa.types.is_large_string(storage.type):
	storage = array_cast(storage, pa.string())
	if pa.types.is_large_string(storage.type):
	try:
	storage = storage.cast(pa.string())
	except pa.ArrowInvalid as e:
	raise ValueError(
	f"Failed to cast large_string to string for Image feature. "
	f"This can happen if string values exceed 2GB. "
	f"Original error: {e}"
	) from e
	if pa.types.is_string(storage.type):

[pdTetteh]

Alright that's helpful

[lhoestq]

Suggested change

	- `pa.string()` - it must contain the "path" data
	- `pa.large_string()` - it must contain the "path" data (will be cast to string if possible)

[lhoestq]

there is a audio_path fixture you can use already

[pdTetteh]

Alright

[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.

[pdTetteh]

Thanks for the review and the suggestions.

I’ll update the PR to:

• add the same large_string handling for Video and Pdf
• update the docstrings to mention pa.large_string()
• switch the audio test to use the existing audio_path fixture
• fix the code quality check failure

[pdTetteh]

I have updated the PR to:

• add the same large_string handling for Video and Pdf
• update the docstrings to mention pa.large_string()
• keep the Audio fix
• add matching regression tests for Audio, Video, and Pdf

I also fixed the code quality issue and reran the targeted feature tests locally.

[lhoestq]

Great ! It looks like you didn't commit the change to switch the audio test to use the existing audio_path fixture and to update the Audio docstring, could you add your changes before we merge ?

[pdTetteh]

Thanks - I updated the audio regression test to use the existing audio fixture available in the test suite and added the missing Audio docstring update for pa.large_string().