Downloaded datasets do not cache at $HF_HOME

[ErfanMoosaviMonazzah]

Describe the bug

In the huggingface course (https://huggingface.co/course/chapter3/2?fw=pt) it said that if we set HF_HOME, downloaded datasets would be cached at specified address but it does not. downloaded models from checkpoint names are downloaded and cached at HF_HOME but this is not the case for datasets, they are still cached at ~/.cache/huggingface/datasets.

Steps to reproduce the bug

Run the following code

from datasets import load_dataset
raw_datasets = load_dataset("glue", "mrpc")
raw_datasets

it downloads and store dataset at ~/.cache/huggingface/datasets

Expected behavior

to cache dataset at HF_HOME.

Environment info

python 3.10.6
Kubuntu 22.04
HF_HOME located on a separate partition

[lhoestq]

Hi ! Can you make sure you set HF_HOME before importing datasets ?

Then you can print

print(datasets.config.HF_CACHE_HOME)
print(datasets.config.HF_DATASETS_CACHE)