Diffusers Environment Setup on Berzelius

This guide explains how to set up the required environment for running diffusers on Berzelius HPC.

Note: As of March 1, 2024, Berzelius supports CUDA 12.2. The following instructions are specifically tailored for compatibility with Python 3.10 and PyTorch 2.5.1 to ensure stable performance.

Prerequisites

• Access to Berzelius HPC
• Loaded Miniforge3 module

Environment Setup

1. Load the required module:

module load Miniforge3

2. Create a new conda environment:

mamba create -n diffusers python=3.10 -y

3. Activate the environment:

mamba activate diffusers

4. Install PyTorch and related packages:

mamba install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=12.1 -c pytorch -c nvidia

5. Install diffusers:

mamba install -c conda-forge diffusers

Usage

After setting up the environment, you can run the training script using:

sbatch run_train_drsk.sh