How to disable caching in load_dataset()?

I am doing the following three steps for a large number of iterations:

  1. Loading a parquet file using load_dataset().
  2. Tokenize it using dataset.map() and HuggingFace tokenizers.
  3. Saving the tokenised dataset on disk in arrow format.

I have the following questions:
a) I don’t want to save anything to the ~/.cache/huggingface/datasets/ as I am saving the final result at a separate location for further use. I tried using load_dataset(..., cache_dir=None) and setting datasets.disable_caching() but none seem to work. From some other threads, I understood that caching can be disabled in dataset.map() and dataset.filter() but not in load_dataset(). How do I disable all types of caching?

b) I plan to train a GPT like transformer model on this tokenised data using the HF ecosystem. I want to conserve disk space but at the same time not make loading extremely slow downstream. What is better arrow vs parquet format in Step 3 above?

@lhoestq @mariosasko @albertvillanova

Hi !

a) I don’t want to save anything to the ~/.cache/huggingface/datasets/ as I am saving the final result at a separate location for further use. I tried using load_dataset(..., cache_dir=None) and setting datasets.disable_caching() but none seem to work. From some other threads, I understood that caching can be disabled in dataset.map() and dataset.filter() but not in load_dataset() . How do I disable all types of caching?

Indeed currently disable_caching uses a temp directory when saving intermediate map results, but load_dataset still writes the original dataset in ~/.cache/huggingface/datasets/

b) I plan to train a GPT like transformer model on this tokenised data using the HF ecosystem. I want to conserve disk space but at the same time not make loading extremely slow downstream. What is better arrow vs parquet format in Step 3 above?

It depends on the dataset size and your training setup, but usually it’s fine using Arrow. For bigger datasets you may use Parquet instead and use streaming=True

I assume someone who is trying to disable the cache would not like the undocumented side effect that load_dataset ignores it. Is this a bug?

I agree it’s kinda confusing though, we might change that at one point and make load_dataset use a temporary directory in that case, though it’s maybe not a trivial change

Btw it’s mentioned here right now: disable_daching() docs