fix: raise error in FolderBasedBuilder when data_dir and data_files are missing

[ArjunJagdale]

Related Issues/PRs

Fixes #6152

---

What changes are proposed in this pull request?

This PR adds a dedicated validation check in the _info() method of the FolderBasedBuilder class to ensure that users provide either data_dir or data_files when loading folder-based datasets (such as audiofolder, imagefolder, etc.).

---

Why this change?

Previously, when calling:

load_dataset("audiofolder")

without specifying data_dir or data_files, the loader would silently fallback to the current working directory, leading to:

• Long loading times
• Unexpected behavior (e.g., scanning unrelated files)

This behavior was discussed in issue #6152. As suggested by maintainers, the fix has now been implemented directly inside the FolderBasedBuilder._info() method — keeping the logic localized to the specific builder instead of a generic loader function.

---

How is this PR tested?

• ✅ Manually tested by calling load_dataset("audiofolder") with no data_dir or data_files → a ValueError is now raised early.
• ✅ Existing functionality (with valid input) remains unaffected.

---

Does this PR require documentation update?

• No

---

Release Notes

Is this a user-facing change?

• Yes

Folder-based datasets now raise an explicit error if neither data_dir nor data_files are specified, preventing unintended fallback to the current working directory.

---

What component(s) does this PR affect?

• area/datasets
• area/load

---

How should the PR be classified?

• rn/bug-fix - A user-facing bug fix

---

Should this be included in the next patch release?

• Yes

[ArjunJagdale]

@lhoestq Moved the logic to FolderBasedBuilder._info() as discussed in previous PR (#7618). Let me know if anything else is needed — happy to update!

[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.

[lhoestq]

lgtm :)