# Accelerate natively compatible with datasets

**URL:** <https://discuss.huggingface.co/t/accelerate-natively-compatible-with-datasets/98343>\
**Category:** 🤗Accelerate\
**Created:** [July 19, 2024, 1:19pm UTC](https://discuss.huggingface.co/t/accelerate-natively-compatible-with-datasets/98343 "2024-07-19T13:19:31Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![mmordig](https://avatars.discourse-cdn.com/v4/letter/m/ad7895/32.png) [@mmordig](https://discuss.huggingface.co/u/mmordig)\
**Post date:** [July 19, 2024, 1:19pm UTC](https://discuss.huggingface.co/t/accelerate-natively-compatible-with-datasets/98343/1 "2024-07-19T13:19:31Z")

</div>

When doing `dataset.map`, does it play nicely with multi-GPU accelerate? I guess `dataset.map` creates a lock file. So do I need to surround it with `main_process_first`?

```auto
with PartialState().main_process_first:
    dataset = dataset.map(...)

```

Is this documented anywhere?