huggingface / huggingface/datasets
Lazy loading of environment variables
- Dominant language
- Python
- Stars
- 22k
- Forks
- 3.4k
- Avg merge
- 5d 7h
- Merged PRs (30d)
- 17
Description
### Describe the bug
Loading a `.env` file after an `import datasets` call does not correctly use the environment variables.
This is due the fact that environment variables are read at import time:
https://github.com/huggingface/datasets/blob/de062f0552a810c52077543c1169c38c1f0c53fc/src/datasets/config.py#L155C1-L155C80
### Steps to reproduce the bug
```bash
# make tmp dir
mkdir -p /tmp/debug-env
# make .env file
echo HF_HOME=/tmp/debug-env/data > /tmp/debug-env/.env
# first load dotenv, downloads to /tmp/debug-env/data
uv run --with datasets,python-dotenv python3 -c \
'import dotenv; dotenv.load_dotenv("/tmp/debug-env/.env"); import datasets; datasets.load_dataset("Anthropic/hh-rlhf")'
# first import datasets, downloads to `~/.cache/huggingface`
uv run --with datasets,python-dotenv python3 -c \
'import datasets; import dotenv; dotenv.load_dotenv("/tmp/debug-env/.env"); datasets.load_dataset("Anthropic/hh-rlhf")'
```
### Expected behavior
I expect that setting environment variables with something like this:
```python3
if __name__ == "__main__":
load_dotenv()
main()
```
works correctly.
### Environment info
"datasets>=3.3.0",
Contributor guide
Assessment
This issue has not been assessed yet.