grokuku / stable-diffusion Goto Github PK

Using web UI 2 (automatic1111) gives an error message about the wrong xFormers version. How should I fix this, other than disabling xformers in the launch parameters? Thanks!

Error message in the web UI:

Error message during first container run:

sd-webui-1  | Launching Web UI with arguments: --listen --port 9000 --enable-insecure-extension-access --medvram --xformers --api
sd-webui-1  | WARNING:xformers:WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
sd-webui-1  |     PyTorch 2.1.2+cu121 with CUDA 1201 (you have 2.2.1+cu121)
sd-webui-1  |     Python  3.11.7 (you have 3.11.8)
sd-webui-1  |   Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
sd-webui-1  |   Memory-efficient attention, SwiGLU, sparse and more won't be available.
sd-webui-1  |   Set XFORMERS_MORE_DETAILS=1 for more details

Below is my docker-compose.yml:

  sd-webui:
    image: holaflenain/stable-diffusion:latest
    restart: unless-stopped
    environment:
      - WEBUI_VERSION=02
      - TZ=America/New_York
      - PUID=1000
      - PGID=1000
    volumes:
      - ./appdata/sd-webui:/config
    ports:
      - 9000:9000
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

Maybe I am missing something here, but I can't start with the --api flag. When I add this to the "extra parameters" in Unraid, the container fails to build.

click histoy shows error
{"detail":"File not allowed: /config/outputs/06-Fooocus/2023-12-30/log.html."}

So, this Docker takes a bit to load up with the Automattic1111 UI option. (I haven't tested the others yet.)

When trying to see why in the logs, I noticed it is uninstalling and reinstalling the torch framework through pip at each startup?

Installing collected packages: triton, torch, torchvision
Attempting uninstall: triton
Found existing installation: triton 2.3.0
Uninstalling triton-2.3.0:
Successfully uninstalled triton-2.3.0
Attempting uninstall: torch
Found existing installation: torch 2.3.0+cu121
Uninstalling torch-2.3.0+cu121:
Successfully uninstalled torch-2.3.0+cu121
Attempting uninstall: torchvision
Found existing installation: torchvision 0.18.1
Uninstalling torchvision-0.18.1:
Successfully uninstalled torchvision-0.18.1

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
xformers 0.0.26.post1 requires torch==2.3.0, but you have torch 2.1.2+cu121 which is incompatible.
Successfully installed torch-2.1.2+cu121 torchvision-0.16.2+cu121 triton-2.1.0

Is this the main culprit to the slow startup times? Why is it doing this and how would I go about fixing it? Or is this expected behavior.

I notice right before this, thee logs are saying it is using the already available 2.3.1 version of torch.

forge runs once and crash at the next run

Please put ffmpeg and the transparent-background, along with the corresponding ckpt_base.pth model, into a Docker container