ImportError: libGL.so.1 on reproducible environnement about generative-models HOT 6 OPEN

I'm also trying to build a docker image. I've tried a couple of base images, including pytorch/pytorch:1.12.1-cuda11.3-cudnn8-runtime which i found here and I always get some dependency error with RUN pip install -r requirements/pt2.txt and they look something like ERROR: No matching distribution found for black==23.7.0 This is my Dockerfile

FROM pytorch/pytorch:1.12.1-cuda11.3-cudnn8-runtime
WORKDIR /app
RUN apt update && apt install -y git
RUN git clone https://github.com/Stability-AI/generative-models.git
WORKDIR /app/generative-models
RUN pip install -r requirements/pt2.txt
RUN pip install .

Edit: this worked for me (not sure about inferencing but I was able to build the app):

FROM pytorch/pytorch:latest
WORKDIR /app
RUN apt update && apt install -y git
RUN git clone https://github.com/Stability-AI/generative-models.git
WORKDIR /app/generative-models
RUN pip install -r requirements/pt2.txt
RUN pip install .

from generative-models.

I have added a Dockerfile with cuda 11.8, xformers, torch 2.0.1 and all required dependencies here

https://github.com/loretoparisi/generative-models

from generative-models.

Hi, you seem to be running into a problem with the opencv python package. It seems most likely that you have some missing dependencies (see here, maybe you can run sudo apt install ffmpeg libsm6 libtext6) or some issue from streamlit (see here, run sudo apt install libgl1 in that case). Try running python -c "import cv2", if that fails, it's the first issue, if that runs through it is caused by streamlit. (Both of these fixes expect that you are running ubuntu)

if this doesn't help, try searching specifically for "opencv python" + your operating system & version.

from generative-models.

Still having the same issue, we might be dealing with a deeper issue related to the Docker container's ability to access the graphics hardware, access the GPU (required by OpenCV for some operations) can be tricky because of the container isolation to host OS, even the Nvidia's Docker images didn't work (maybe due to the limitation of graphics ability), I think this has to be run outside of containers. How do you run this?

from generative-models.

See here for installation instructions. I know that someone manged to make it work in docker, but for your nvidia/docker issues I recommend opening an issue with one of them.

from generative-models.

See here for installation instructions. I know that someone manged to make it work in docker, but for your nvidia/docker issues I recommend opening an issue with one of them.

I confirm I'm having the same issue in Docker

root@4ff0cee5b260:/opt/program# python scripts/demo/detect.py assets/test_image.png 
Traceback (most recent call last):
  File "/opt/program/scripts/demo/detect.py", line 3, in <module>
    import cv2
  File "/usr/local/lib/python3.10/dist-packages/cv2/__init__.py", line 181, in <module>
    bootstrap()
  File "/usr/local/lib/python3.10/dist-packages/cv2/__init__.py", line 153, in bootstrap
    native_module = importlib.import_module("cv2")
  File "/usr/lib/python3.10/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
ImportError: libGL.so.1: cannot open shared object file: No such file or directory
root@4ff0cee5b260:/opt/program# 

In My docker file I have

FROM pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.0-cuda11.8.0

# ...

COPY requirements/pt2.txt /opt/program/requirements.txt
# install requirements in order
RUN pip3 install -r /opt/program/requirements.txt

from generative-models.