Hugging Face datasets "trust_remote_code" required in CI

load_dataset found a custom loading or processing script in the dataset repo. Newer datasets refuses to run repo-provided code unless you explicitly pass trust_remote_code=True, so the load aborts.

What this error means

load_dataset fails with "ValueError: The repository for X contains custom code which must be executed to correctly load the dataset. ... Please pass the argument trust_remote_code=True."

ValueError: The repository for squad contains custom code which must be executed to
correctly load the dataset. You can inspect the repository content at https://hf.co/datasets/...
Please pass the argument `trust_remote_code=True` to allow custom code to be run.

Common causes

The dataset ships a loading script

Script-based datasets execute repo code to build splits. Recent datasets versions block that by default for safety.

A datasets upgrade changed the default

Older code that loaded the dataset fine now fails because the library tightened the default to not trust remote code.

How to fix it

Opt in after reviewing the script

Inspect the dataset repo, then pass trust_remote_code=True if you accept running its code.

from datasets import load_dataset
ds = load_dataset("some/dataset", trust_remote_code=True)

Prefer a no-script dataset format

Use a dataset stored as Parquet/CSV/JSON (no loading script) so no remote code execution is needed.

ds = load_dataset("parquet", data_files="data/*.parquet")

How to prevent it

• Prefer script-free dataset formats in CI.
• Pin the dataset revision so its script does not change under you.
• Set trust_remote_code consciously after reviewing the repo.

Related guides