# Distributed training on just cpu on a single node

**URL:** <https://discuss.huggingface.co/t/distributed-training-on-just-cpu-on-a-single-node/62970>\
**Category:** 🤗Transformers\
**Created:** [November 21, 2023, 6:55am UTC](https://discuss.huggingface.co/t/distributed-training-on-just-cpu-on-a-single-node/62970 "2023-11-21T06:55:07Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![ruta-8968](https://avatars.discourse-cdn.com/v4/letter/r/b9e5f3/32.png) [@ruta-8968](https://discuss.huggingface.co/u/ruta-8968)\
**Post date:** [November 21, 2023, 6:55am UTC](https://discuss.huggingface.co/t/distributed-training-on-just-cpu-on-a-single-node/62970/1 "2023-11-21T06:55:07Z")

</div>

How to use ddp\_backend=“gloo” for trainer for running distributed training on multiple docker containers on a single node?

1. Do I still need to initialize with dist.init\_process\_group within the script?
2. What would be the world size while running multiple containers on a single node?