Install stable-diffusion-webui with ROCm support on Linux

installation.md

Install stable-diffusion-webui with ROCm support on Arch Linux

First time installation:

1. Clone and switch to the stable-diffusion-webui folder

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui

2. Create a new python environment and switch to it

python -m venv venv
source venv/bin/activate

3. Update pip and wheel

python -m pip install --upgrade pip wheel

4. To prevent a memory leak in the application, we need to preload /usr/lib/libtcmalloc.so, this is part of the gperftools package. So let's install that:

sudo pacman -S gperftools

5. Create and edit a new shell file which we will use to start stable-diffusion-webui

touch start.sh
nano start.sh

6. Add this content to the file and save it with CTRL+O and exit with CTRL+X

#!/usr/bin/env bash
source venv/bin/activate
export HSA_OVERRIDE_GFX_VERSION=10.3.0
export MIOPEN_DEBUG_COMGR_HIP_PCH_ENFORCE=0
export PYTORCH_HIP_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
export PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
export TORCH_COMMAND='pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm5.7'
export REQS_FILE='requirements.txt'
export LD_PRELOAD=/usr/lib/libtcmalloc.so
python launch.py --precision full --no-half --enable-insecure-extension-access --listen --opt-sub-quad-attention --no-half-vae --disable-nan-check --medvram

7. Make the file executable

chmod +x start.sh

8. Deactivate the python environment

deactivate

Running stable-diffusion-webui

Switch to the stable-diffusion-webui folder and run:

./start.sh

Hello, any idea ?

This seems to be an issue with stable-diffusion-web itself. It also happens with their own runner. For some reason, after trying several times, it eventually worked. No idea what exactly is causing it.