Some Docker Issues

[ghost]

I was having some issues getting Docker running with Cuda on Ubuntu. Not sure if that file is being maintained frequently but take a look what worked for me, feel free to commit. You'd need to run this as a replacement for the original Dockerfile, not as an upgrade.

Start with an official Python base image

FROM python:3.11.6

Avoid interactive prompts from apt (e.g., tzdata)

ENV DEBIAN_FRONTEND=noninteractive

Install necessary system dependencies

RUN apt-get update && apt-get -y upgrade &&
apt-get install -y bash sudo apt-utils git gcc perl make ffmpeg openssl curl

Setup user biniou and use it to install

RUN adduser --disabled-password --gecos '' biniou
USER biniou
WORKDIR /home/biniou

Pull repo

RUN git clone https://github.com/Woolverine94/biniou.git biniou
WORKDIR /home/biniou/biniou

Create a virtual environment and install Python packages

RUN python -m venv env
SHELL ["/bin/bash", "-c"]
RUN source env/bin/activate &&
pip install -U pip wheel &&
pip install -U torch==2.1.0 torchvision torchaudio &&
FORCE_CMAKE=1 pip install -U llama-cpp-python &&
pip install -r requirements.txt

Install biniou using the provided install script

RUN ./install.sh

Set up directories and permissions

RUN mkdir -p /home/biniou/.cache/huggingface /home/biniou/biniou/gfpgan
RUN chmod +x /home/biniou/biniou/webui.sh

Exposing port 7860

EXPOSE 7860/tcp

Launch at startup

CMD [ "/home/biniou/biniou/webui.sh" ]

Reset DEBIAN_FRONTEND for interactive use when running the container

ENV DEBIAN_FRONTEND=dialog

[Woolverine94]

Hello @AndrewDigitalChaos,

Thanks for your feedback.

Default biniou installation is natively CPU-only (and it will stay that way), so it don't work out-of-the-box with CUDA in a docker env.

Since commit da3be41 of yesterday, you should be able to activate CUDA directly from biniou's webUI control panel, even in docker.
Unfortunately, I don't have access to CUDA-compatible hardware to validate it's working, even if I can confirm that replacement of pyTorch flavor is done by the update script.

Anyway, this is more a workaround for using CUDA in docker than a turnkey CUDA docker installation.

Thanks A LOT for your contribution, I will think about a way later today to introduce your changes (probably via a specific dockerfile).

[Woolverine94]

@AndrewDigitalChaos,

Following your ideas, I've added a specific CUDA Dockerfile that can be used as following to build a docker image :

docker build -t biniou https://raw.githubusercontent.com/Woolverine94/biniou/main/CUDA/Dockerfile

Thanks again for your contribution to biniou.