Docker image

[zecloud]

Hi,
I've made a docker image to use it with Gpu on Azure with Azure Container instance

[snpranav]

Hey @zecloud , thanks for adding the Dockerfile. I tested it out on my Ubuntu rig and it worked.
Here are a couple of changes I'd suggest:

• In the Dockerfile, please make DEBIAN_FRONTEND=noninteractive for all commands since some of the python packages require the user input during the package setup process, so the build usually fails.
  So just add ARG DEBIAN_FRONTEND=noninteractive to the top of the file.

• It'd be nice to also make this change to the Readme file as to instructions on how to run this after mounting GPUs to the docker container.
  So something like -> docker run --gpus all tortoise-tts:latest

But otherwise, this PR works!

[zecloud]

Hey @snpranav
thanks for your suggestion, i already had DEBIAN_FRONTEND=noninteractive at the top of the image and i can't remember why i made an ugly quickfix for the libsndfile1-dev. So fix is fixed.
And i've added a little documentation for the docker image.
Let me know if that's work for you
Best !

[kfatehi]

Hey @zecloud I was having errors running on my 4090 with those older versions of cuda, but this is working well:

FROM nvidia/cuda:11.8.0-base-ubuntu20.04

ENV PYTHON_VERSION=3.8

RUN export DEBIAN_FRONTEND=noninteractive \
    && apt-get -qq update \
    && apt-get -qq install --no-install-recommends \
    libsndfile1-dev \
    git \
    python${PYTHON_VERSION} \
    python${PYTHON_VERSION}-venv \
    python3-pip \
    && rm -rf /var/lib/apt/lists/*

RUN ln -s -f /usr/bin/python${PYTHON_VERSION} /usr/bin/python3 && \
    ln -s -f /usr/bin/python${PYTHON_VERSION} /usr/bin/python && \
    ln -s -f /usr/bin/pip3 /usr/bin/pip

RUN pip install --upgrade pip

# 2. Copy files
COPY . /src

RUN pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu118

WORKDIR /src
# 3. Install dependencies
RUN pip install -r requirements-docker.txt

RUN python3 setup.py install

[zecloud]

Hey @kfatehi,
Thanks for your suggestion, i didn't have a 4090 to test it.
So i've added your dockerfile to my repo as a new Dockerfile.
I'll test it when i can on an older GPU with updated cuda.

[bmurray]

This works for me, but is the desired behaviour for it to download several gigs of model data every time the container runs? I feel theres a bug here somewhere. Even with a persistent models dir, and TORTOISE_MODELS_DIR set, it spends about half the time downloading.

[bmurray]

After some debugging, it looks like its not really the Docker image thats the issue, but how Huggingface caches the transformers. I solved it by mounting a volume to /cache and using the environment variable.

TRANSFORMERS_CACHE=/cache

[zecloud]

hey @bmurray
Thanks for noticing, in my deployment i had an environment variable to cache the transformers, but i forgot to put it in my Readme. I've updated the readme with it.