# Monai Label defaulting to CPU only training with segmentation model

**URL:** <https://discourse.slicer.org/t/monai-label-defaulting-to-cpu-only-training-with-segmentation-model/40301>\
**Category:** Support\
**Tags:** segmentation\
**Created:** [November 21, 2024, 4:07am UTC](https://discourse.slicer.org/t/monai-label-defaulting-to-cpu-only-training-with-segmentation-model/40301 "2024-11-21T04:07:30Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Saim\_ali](https://sea2.discourse-cdn.com/flex002/user_avatar/discourse.slicer.org/saim_ali/32/66983_2.png) [@Saim\_ali](https://discourse.slicer.org/u/Saim_ali)\
**Post date:** [November 21, 2024, 4:07am UTC](https://discourse.slicer.org/t/monai-label-defaulting-to-cpu-only-training-with-segmentation-model/40301/1 "2024-11-21T04:07:30Z")

</div>

Hi all, novice in programming so I am learning as I go.

I have a set of CT scans with custom labels and I have figured out how to employ Monai Label to customize the segmentation model but I am having some trouble using CUDA for training. I believe I have a capable GPU (RTX 2070 super), I checked nvidia-smi in terminal and it says CUDA 12.7.

I installed PyTorch 2.3.1 w/ Cuda 11.8 in my python environment before installing monai label with Pip but the monai label install uninstalls torch 2.3.1/cuda11.8 and installs 2.5.1 instead.

I am now training the segmentation model with my dataset but it is using cpu and taking quite a long time.

TLDR: I have CT scans + labels but can’t figure out how properly use monai label with cuda