huggingface / huggingface/datasets
Using disk for memory with the method `from_dict`
- Dominant language
- Python
- Stars
- 22k
- Forks
- 3.4k
- Avg merge
- 5d 7h
- Merged PRs (30d)
- 17
Description
**Is your feature request related to a problem? Please describe.**
I start with an empty dataset. In a loop, at each iteration, I create a new dataset with the method `from_dict` (based on some data I load) and I concatenate this new dataset with the one at the previous iteration. After some iterations, I have an OOM error.
**Describe the solution you'd like**
The method `from_dict` loads the data in RAM. It could be good to add an option to use the disk instead.
**Describe alternatives you've considered**
To solve the problem, I have to do an intermediate step where I save the new datasets at each iteration with `save_to_disk`. Once it's done, I open them all and concatenate them.
Contributor guide
Assessment
This issue has not been assessed yet.