Is CPU-Only Usage Possible? about stable-diffusion-webui-docker

[GeorgiaM-honestly]

I'm running in a VM but i didn't expect performance to be THIS bad. (I'm not complaining about the project). Having a blast :)

Prompt: spleens made of cheese, f/1.4, 50mm, high detail, photo realistic, fujifilm tri-x, ISO 6400
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 2343712771, size: 512x512
Time taken: 109.44s

Another:

Prompt: surreal, wading into the ocean with fishing poles as legs, f/1.4, 50mm, high detail, photo realistic, fujifilm tri-x, ISO 6400
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 845967881, size: 512x512
Time taken: 107.21s

[AbdBarho]

That is a really good question, I will do some testing and report back

[GeorgiaM-honestly]

Okay. I got it to start though it complains:

 ⠿ Container stable-diffusion-webui-docker-model-1  Running                                                                                                                                         0.0s
Attaching to stable-diffusion-webui-docker-model-1
stable-diffusion-webui-docker-model-1  | Global Step: 470000
stable-diffusion-webui-docker-model-1  | UNet: Running in eps-prediction mode
stable-diffusion-webui-docker-model-1  | Traceback (most recent call last):
stable-diffusion-webui-docker-model-1  |   File "scripts/webui.py", line 449, in <module>
stable-diffusion-webui-docker-model-1  |     model,modelCS,modelFS,device, config = load_SD_model()
stable-diffusion-webui-docker-model-1  |   File "scripts/webui.py", line 420, in load_SD_model
stable-diffusion-webui-docker-model-1  |     model.cuda()
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/pytorch_lightning/core/mixins/device_dtype_mixin.py", line 127, in cuda
stable-diffusion-webui-docker-model-1  |     return super().cuda(device=device)
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 688, in cuda
stable-diffusion-webui-docker-model-1  |     return self._apply(lambda t: t.cuda(device))
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 578, in _apply
stable-diffusion-webui-docker-model-1  |     module._apply(fn)
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 578, in _apply
stable-diffusion-webui-docker-model-1  |     module._apply(fn)
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 578, in _apply
stable-diffusion-webui-docker-model-1  |     module._apply(fn)
stable-diffusion-webui-docker-model-1  |   [Previous line repeated 1 more time]
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 601, in _apply
stable-diffusion-webui-docker-model-1  |     param_applied = fn(param)
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 688, in <lambda>
stable-diffusion-webui-docker-model-1  |     return self._apply(lambda t: t.cuda(device))
stable-diffusion-webui-docker-model-1  |   File "/opt/conda/lib/python3.8/site-packages/torch/cuda/__init__.py", line 216, in _lazy_init
stable-diffusion-webui-docker-model-1  |     torch._C._cuda_init()
stable-diffusion-webui-docker-model-1  | RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx
stable-diffusion-webui-docker-model-1 exited with code 1

Therefore I looked at the CLI flags and changed the yaml config to:

version: '3.9'

services:
  model:
    build:
      context: ./hlky/
      args:
        # You can choose any commit sha from https://github.com/hlky/stable-diffusion/commits/main
        # USE AT YOUR OWN RISK! otherwise just leave it empty.
        WEBUI_SHA:
    restart: on-failure
    ports:
      - "7860:7860"
    volumes:
      - ./cache:/cache
      - ./output:/output
      - ./models:/models
    environment:
# Default
#      - CLI_ARGS=--extra-models-cpu --optimized-turbo
# https://github.com/hlky/stable-diffusion/blob/bb765f1897c968495ffe12a06b421d97b56d5ae1/scripts/webui.py
      - CLI_ARGS=--extra-models-cpu --optimized-turbo --esrgan-cpu --gfpgan-cpu
#    deploy:
#      resources:
#        reservations:
#          devices:
#              - driver: nvidia
#                device_ids: ['0']
#                capabilities: [gpu]

Now, despite the error, the container IS running and taking up some resources. However when I try to connect to 127.0.0.1:7860 from the system the docker is running on, I get connection refused:

nc 127.0.0.1 7860
(UNKNOWN) [127.0.0.1] 7860 (?) : Connection refused

Which is curious, because it is listening:

netstat -anp | grep :7860
tcp        0      0 0.0.0.0:7860            0.0.0.0:*               LISTEN      3679/docker-proxy   
tcp6       0      0 :::7860                 :::*                    LISTEN      3685/docker-proxy   

Maybe i'm missing something :P

EDIT:

I now see why the container is running and using resources; it is restarting itself repeatedly.

[AbdBarho]

@GeorgiaM-honestly you can stop the container with docker compose down

I am still looking into the cpu stuff

[AbdBarho]

@GeorgiaM-honestly here is how to run it on cpu

cd AUTOMATIC1111

update the docker compose file

version: '3.9'

services:
  model:
    build: .
    ports:
      - "7860:7860"
    volumes:
      - ../cache:/cache
      - ../output:/output
      - ../models:/models
    environment:
      - CLI_ARGS=--no-half --precision full

then

docker compose up --build

I only tested 64x64, but it seems to work.

[GeorgiaM-honestly]

Okay. I no longer get errors as before, and 127.0.0.1 7860 isn't rejecting connections now however it isn't sending anything either and closes the connection. The container is restarting itself.

is docker compose up --build supposed to be run in AUTOMATIC1111/ or the root of the repo? I did the later.

[AbdBarho]

no it should be in the AUTOMATIC1111 folder, because that is a different container

it will take a while to build, but its worth it, its much more stable than the main one.

[GeorgiaM-honestly]

It worked!! Default settings which was 512x512 on 32x4.1GHz cpus in a VM. What else can I do to help you test?

[AbdBarho]

@GeorgiaM-honestly thank you for your offer!

How much time did it take for that image size? just to get a feeling for it. On my machine 4 cores / 8 threads / 2.4ghz / windows 10 / 20 steps / 64 x 64 it took around 8 seconds.

I will close this issue now, feel free to open a new one if you face any problems or you want to request a new feature.

Otherwise, Enjoy!