1
0
Fork 0

optimize args

This commit is contained in:
Massaki Archambault 2023-08-03 22:58:31 -04:00
parent ce324b577a
commit 949a247642
2 changed files with 12 additions and 4 deletions

View File

@ -1,5 +1,9 @@
#!/bin/bash
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui /shared/stable-diffusion-webui
cd /shared
ln -s ./stable-diffusion-webui/models models
ln -s ./stable-diffusion-webui/outputs outputs
ln -s ./stable-diffusion-webui/embeddings embeddings
cd /shared/stable-diffusion-webui
exec ./webui.sh

View File

@ -3,11 +3,15 @@ services:
stable-diffusion:
build: .
environment:
# for 5700xt compat
# ref: https://old.reddit.com/r/StableDiffusion/comments/ww436j/howto_stable_diffusion_on_an_amd_gpu/imcpmo8/
- HSA_OVERRIDE_GFX_VERSION=10.3.0
- COMMANDLINE_ARGS=--precision full --no-half
network_mode: host
# for 5700xt compat
# ref: https://old.reddit.com/r/StableDiffusion/comments/ww436j/howto_stable_diffusion_on_an_amd_gpu/imcpmo8/
- PYTORCH_HIP_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
- COMMANDLINE_ARGS=--listen --theme=dark --precision full --no-half
# - COMMANDLINE_ARGS=--listen --theme=dark --precision full --no-half --medvram
# - COMMANDLINE_ARGS=--listen --theme=dark --use-cpu all --precision full --no-half --skip-torch-cuda-test
ports:
- 7860:7860
devices:
- /dev/kfd
- /dev/dri