Docker support (Tested on RTX3080) + ignoring reproducable files / cache

.dockerignore

@@ -0,0 +1,6 @@
+models/
+Dockerfile
+README.md
+LICENSE
+assets/
+*.egg-info

.gitignore

@@ -0,0 +1,5 @@
+models/ldm/text2img-large/
+outputs/
+src/
+__pycache__/
+*.egg-info

Dockerfile

@@ -0,0 +1,38 @@
+FROM nvidia/cuda:11.3.1-cudnn8-devel-ubuntu20.04
+MAINTAINER Peter Willemsen <peter@codebuffet.co>
+RUN echo "Installing dependencies..." && \
+    apt-get update && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y curl wget sudo git build-essential cmake pkg-config liblzma-dev libbz2-dev zlib1g-dev libssl-dev zsh clang && \
+    apt-get dist-upgrade -y && \
+	rm -rf /var/lib/apt/lists/*
+
+WORKDIR /src/python
+RUN wget https://www.python.org/ftp/python/3.8.5/Python-3.8.5.tgz -O python-src.tar.gz && \
+    tar xzvf python-src.tar.gz --strip-components=1 && \
+    rm python-src.tar.gz && \
+    ./configure --enable-optimizations --prefix=/opt/python-3.8.5 && \
+    make && \
+    make install && \
+    rm -rf /src/python
+WORKDIR / 
+ENV PATH="/opt/python-3.8.5/bin:${PATH}"
+
+RUN python3 -m pip install pip==20.3
+RUN pip3 install torch==1.10.1+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
+RUN pip3 install numpy==1.19.2 torchvision==0.11.2 albumentations==0.4.3 opencv-python==4.1.2.30 pudb==2019.2 imageio==2.9.0 imageio-ffmpeg==0.4.2 pytorch-lightning==1.6.1 omegaconf==2.1.1 test-tube>=0.7.5 streamlit>=0.73.1 einops==0.3.0 torch-fidelity==0.3.0 transformers==4.3.1 -e "git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers" -e "git+https://github.com/openai/CLIP.git@main#egg=clip"
+
+RUN mkdir -p /opt/ldm_package
+ADD ./setup.py /opt/ldm_package
+ADD ./ldm /opt/ldm_package/ldm
+ADD ./configs /opt/ldm_package/configs
+RUN pip3 install -e /opt/ldm_package
+
+WORKDIR /opt/ldm
+
+# Add dev user
+RUN useradd -ms /bin/zsh ldm-dev && \
+    usermod -aG sudo ldm-dev && \
+    echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers
+USER ldm-dev
+
+ENTRYPOINT ["python3"]

README.md

@@ -58,6 +58,20 @@ As a rule of thumb, higher values of `scale` produce better samples at the cost
 Furthermore, increasing `ddim_steps` generally also gives higher quality samples, but returns are diminishing for values > 250.
 Fast sampling (i.e. low values of `ddim_steps`) while retaining good quality can be achieved by using `--ddim_eta 0.0`.  
 
+## Installing on Docker
+
+- Note: depending on CUDA/GPU version, you might have to change the first line in te Dockerfile or change the torch version being installed. 
+- Build the image: `docker build . --tag latent-diffusion`
+- For text-to-image, download the pre-trained weights (5.7GB):
+  ```
+  mkdir -p models/ldm/text2img-large/
+  wget -O models/ldm/text2img-large/model.ckpt https://ommer-lab.com/files/latent-diffusion/nitro/txt2img-f8-large/model.ckpt
+  ```
+- Sample with (Make sure to call in the directory of this repo):
+  ```
+  docker run --name=tmp-diffusion --rm --gpus all -it -v "$(pwd):/opt/ldm" latent-diffusion /opt/ldm/scripts/txt2img.py --prompt "A large blue whale on a freight ship, vector art" --ddim_eta 0.0 --n_samples 4 --n_iter 4 --scale 5.0 --ddim_steps 50
+  ```
+
 #### Beyond 256²
 
 For certain inputs, simply running the model in a convolutional fashion on larger features than it was trained on