Refactor invoke (#405)

Fixes a problem with cross attention class missing from diffusers


models are now taken from the huggingFace cache.


50eb02f68b

data/.gitignore

@@ -19,3 +19,6 @@
 /Hypernetworks
 /VAE
 /embeddings
+/Lora
+/ControlNet
+/openpose

docker-compose.yml

@@ -6,6 +6,7 @@ x-base_service: &base_service
     volumes:
       - &v1 ./data:/data
       - &v2 ./output:/output
+    stop_signal: SIGINT
     deploy:
       resources:
         reservations:
@@ -27,7 +28,7 @@ services:
     <<: *base_service
     profiles: ["auto"]
     build: ./services/AUTOMATIC1111
-    image: sd-auto:32
+    image: sd-auto:51
     environment:
       - CLI_ARGS=--allow-code --medvram --xformers --enable-insecure-extension-access --api
 
@@ -36,16 +37,16 @@ services:
     profiles: ["auto-cpu"]
     deploy: {}
     environment:
-      - CLI_ARGS=--no-half --precision full
+      - CLI_ARGS=--no-half --precision full --allow-code --enable-insecure-extension-access --api
 
   invoke:
     <<: *base_service
     profiles: ["invoke"]
     build: ./services/invoke/
-    image: sd-invoke:17
+    image: sd-invoke:27
     environment:
       - PRELOAD=true
-      - CLI_ARGS=
+      - CLI_ARGS=--no-nsfw_checker --no-safety_checker --xformers
 
 
   sygil: &sygil

services/AUTOMATIC1111/Dockerfile

@@ -1,14 +1,6 @@
-# syntax=docker/dockerfile:1
-
 FROM alpine/git:2.36.2 as download
 
-SHELL ["/bin/sh", "-ceuxo", "pipefail"]
-
-RUN <<EOF
-cat <<'EOE' > /clone.sh
-mkdir -p repositories/"$1" && cd repositories/"$1" && git init && git remote add origin "$2" && git fetch origin "$3" --depth=1 && git reset --hard "$3" && rm -rf .git
-EOE
-EOF
+COPY clone.sh /clone.sh
 
 RUN . /clone.sh taming-transformers https://github.com/CompVis/taming-transformers.git 24268930bf1dce879235a7fddd0b2355b84d7ea6 \
   && rm -rf data assets **/*.ipynb
@@ -26,25 +18,23 @@ RUN . /clone.sh clip-interrogator https://github.com/pharmapsychotic/clip-interr
 
 FROM alpine:3.17 as xformers
 RUN apk add --no-cache aria2
-RUN aria2c -x 5 --dir / --out wheel.whl 'https://github.com/AbdBarho/stable-diffusion-webui-docker/releases/download/4.1.0/xformers-0.0.16.dev421-cp310-cp310-manylinux2014_x86_64.whl'
+RUN aria2c -x 5 --dir / --out wheel.whl 'https://github.com/AbdBarho/stable-diffusion-webui-docker/releases/download/5.0.0/xformers-0.0.17.dev449-cp310-cp310-manylinux2014_x86_64.whl'
 
 FROM python:3.10.9-slim
 
-SHELL ["/bin/bash", "-ceuxo", "pipefail"]
-
 ENV DEBIAN_FRONTEND=noninteractive PIP_PREFER_BINARY=1
 
-RUN PIP_NO_CACHE_DIR=1 pip install torch==1.13.1+cu117 torchvision --extra-index-url https://download.pytorch.org/whl/cu117
+RUN --mount=type=cache,target=/root/.cache/pip \
+  pip install torch==1.13.1+cu117 torchvision --extra-index-url https://download.pytorch.org/whl/cu117
 
 RUN apt-get update && apt install fonts-dejavu-core rsync git jq moreutils -y && apt-get clean
 
 
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
-cd stable-diffusion-webui
-git reset --hard d7aec59c4eb02f723b3d55c6f927a42e97acd679
-pip install -r requirements_versions.txt
-EOF
+RUN --mount=type=cache,target=/root/.cache/pip \
+  git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git && \
+  cd stable-diffusion-webui && \
+  git reset --hard d7aec59c4eb02f723b3d55c6f927a42e97acd679 && \
+  pip install -r requirements_versions.txt
 
 RUN --mount=type=cache,target=/root/.cache/pip  \
   --mount=type=bind,from=xformers,source=/wheel.whl,target=/xformers-0.0.15-cp310-cp310-linux_x86_64.whl \
@@ -53,41 +43,45 @@ RUN --mount=type=cache,target=/root/.cache/pip  \
 ENV ROOT=/stable-diffusion-webui
 
 
-COPY --from=download /git/ ${ROOT}
+COPY --from=download /repositories/ ${ROOT}/repositories/
 RUN mkdir ${ROOT}/interrogate && cp ${ROOT}/repositories/clip-interrogator/data/* ${ROOT}/interrogate
 RUN --mount=type=cache,target=/root/.cache/pip \
   pip install -r ${ROOT}/repositories/CodeFormer/requirements.txt
 
 RUN --mount=type=cache,target=/root/.cache/pip \
-  pip install opencv-python-headless \
+  pip install pyngrok \
   git+https://github.com/TencentARC/GFPGAN.git@8d2447a2d918f8eba5a4a01463fd48e45126a379 \
   git+https://github.com/openai/CLIP.git@d50d76daa670286dd6cacf3bcd80b5e4823fc8e1 \
-  git+https://github.com/mlfoundations/open_clip.git@bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b \
-  pyngrok
+  git+https://github.com/mlfoundations/open_clip.git@bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b
 
 # Note: don't update the sha of previous versions because the install will take forever
 # instead, update the repo state in a later step
 
-ARG SHA=82725f0ac439f7e3b67858d55900e95330bbd326
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-cd stable-diffusion-webui
-git fetch
-git reset --hard ${SHA}
-pip install -r requirements_versions.txt
-EOF
+# TODO: either remove if fixed in A1111 (unlikely) or move to the top with other apt stuff
+RUN apt-get -y install libgoogle-perftools-dev && apt-get clean
+ENV LD_PRELOAD=libtcmalloc.so
 
+ARG SHA=a9fed7c364061ae6efb37f797b6b522cb3cf7aa2
 RUN --mount=type=cache,target=/root/.cache/pip \
-  pip install -U opencv-python-headless transformers>=4.24
+  cd stable-diffusion-webui && \
+  git fetch && \
+  git reset --hard ${SHA} && \
+  pip install -r requirements_versions.txt
+
+RUN --mount=type=cache,target=/root/.cache/pip  pip install -U opencv-python-headless
 
 COPY . /docker
 
-RUN <<EOF
-python3 /docker/info.py ${ROOT}/modules/ui.py
-mv ${ROOT}/style.css ${ROOT}/user.css
-EOF
+RUN \
+  python3 /docker/info.py ${ROOT}/modules/ui.py && \
+  mv ${ROOT}/style.css ${ROOT}/user.css && \
+  # one of the ugliest hacks I ever wrote \
+  sed -i 's/in_app_dir = .*/in_app_dir = True/g' /usr/local/lib/python3.10/site-packages/gradio/routes.py
 
 WORKDIR ${ROOT}
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_VISIBLE_DEVICES=all
 ENV CLI_ARGS=""
 EXPOSE 7860
 ENTRYPOINT ["/docker/entrypoint.sh"]
-CMD python3 -u webui.py --listen --port 7860 ${CLI_ARGS}
+CMD python -u webui.py --listen --port 7860 ${CLI_ARGS}

services/AUTOMATIC1111/clone.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+set -Eeuox pipefail
+
+mkdir -p /repositories/"$1"
+cd /repositories/"$1"
+git init
+git remote add origin "$2"
+git fetch origin "$3" --depth=1
+git reset --hard "$3"
+rm -rf .git

services/AUTOMATIC1111/entrypoint.sh

@@ -4,6 +4,10 @@ set -Eeuo pipefail
 
 # TODO: move all mkdir -p ?
 mkdir -p /data/config/auto/scripts/
+# mount scripts individually
+find "${ROOT}/scripts/" -maxdepth 1 -type l -delete
+cp -vrfTs /data/config/auto/scripts/ "${ROOT}/scripts/"
+
 cp -n /docker/config.json /data/config/auto/config.json
 jq '. * input' /data/config/auto/config.json /docker/config.json | sponge /data/config/auto/config.json
 
@@ -11,9 +15,6 @@ if [ ! -f /data/config/auto/ui-config.json ]; then
   echo '{}' >/data/config/auto/ui-config.json
 fi
 
-# copy scripts, we cannot just mount the directory because it will override the already provided scripts in the repo
-cp -rfT /data/config/auto/scripts/ "${ROOT}/scripts"
-
 declare -A MOUNTS
 
 MOUNTS["/root/.cache"]="/data/.cache"
@@ -33,6 +34,9 @@ MOUNTS["${ROOT}/models/hypernetworks"]="/data/Hypernetworks"
 MOUNTS["${ROOT}/models/torch_deepdanbooru"]="/data/Deepdanbooru"
 MOUNTS["${ROOT}/models/BLIP"]="/data/BLIP"
 MOUNTS["${ROOT}/models/midas"]="/data/MiDaS"
+MOUNTS["${ROOT}/models/Lora"]="/data/Lora"
+MOUNTS["${ROOT}/models/ControlNet"]="/data/ControlNet"
+MOUNTS["${ROOT}/models/openpose"]="/data/openpose"
 
 MOUNTS["${ROOT}/embeddings"]="/data/embeddings"
 MOUNTS["${ROOT}/config.json"]="/data/config/auto/config.json"

services/download/download.sh

@@ -3,7 +3,7 @@
 set -Eeuo pipefail
 
 # TODO: maybe just use the .gitignore file to create all of these
-mkdir -vp /data/.cache /data/StableDiffusion /data/Codeformer /data/GFPGAN /data/ESRGAN /data/BSRGAN /data/RealESRGAN /data/SwinIR /data/LDSR /data/ScuNET /data/embeddings /data/VAE /data/Deepdanbooru /data/MiDaS
+mkdir -vp /data/.cache /data/StableDiffusion /data/Codeformer /data/GFPGAN /data/ESRGAN /data/BSRGAN /data/RealESRGAN /data/SwinIR /data/LDSR /data/ScuNET /data/embeddings /data/VAE /data/Deepdanbooru /data/MiDaS /data/Lora /data/ControlNet /data/openpose
 
 echo "Downloading, this might take a while..."
 
@@ -15,10 +15,11 @@ parallel --will-cite -a /docker/checksums.sha256 "echo -n {} | sha256sum -c"
 
 cat <<EOF
 By using this software, you agree to the following licenses:
-https://github.com/CompVis/stable-diffusion/blob/main/LICENSE
 https://github.com/AbdBarho/stable-diffusion-webui-docker/blob/master/LICENSE
-https://github.com/sd-webui/stable-diffusion-webui/blob/master/LICENSE
+https://github.com/CompVis/stable-diffusion/blob/main/LICENSE
+https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/LICENSE.txt
 https://github.com/invoke-ai/InvokeAI/blob/main/LICENSE
+https://github.com/sd-webui/stable-diffusion-webui/blob/master/LICENSE
 https://github.com/cszn/BSRGAN/blob/main/LICENSE
 https://github.com/sczhou/CodeFormer/blob/master/LICENSE
 https://github.com/TencentARC/GFPGAN/blob/master/LICENSE

services/invoke/Dockerfile

@@ -1,58 +1,58 @@
-# syntax=docker/dockerfile:1
+FROM alpine:3.17 as xformers
+RUN apk add --no-cache aria2
+RUN aria2c -x 5 --dir / --out wheel.whl 'https://github.com/AbdBarho/stable-diffusion-webui-docker/releases/download/5.0.0/xformers-0.0.17.dev449-cp310-cp310-manylinux2014_x86_64.whl'
+
+
 
 FROM python:3.10-slim
-SHELL ["/bin/bash", "-ceuxo", "pipefail"]
 
 ENV DEBIAN_FRONTEND=noninteractive PIP_EXISTS_ACTION=w  PIP_PREFER_BINARY=1
 
 
-RUN --mount=type=cache,target=/root/.cache/pip \
-  pip install torch==1.12.0+cu116 --extra-index-url https://download.pytorch.org/whl/cu116
-
-RUN apt-get update && apt-get install git -y && apt-get clean
-
-RUN git clone https://github.com/invoke-ai/InvokeAI.git /stable-diffusion
-
-WORKDIR /stable-diffusion
-
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-git reset --hard 5c31feb3a1096d437c94b6e1c3224eb7a7224a85
-git config --global http.postBuffer 1048576000
-pip install -r binary_installer/py3.10-linux-x86_64-cuda-reqs.txt
-EOF
+RUN --mount=type=cache,target=/root/.cache/pip pip install torch==1.13.1+cu117 torchvision --extra-index-url https://download.pytorch.org/whl/cu117
 
 
 # patch match:
 # https://github.com/invoke-ai/InvokeAI/blob/main/docs/installation/INSTALL_PATCHMATCH.md
-RUN <<EOF
-apt-get update
-# apt-get install build-essential python3-opencv libopencv-dev -y
-apt-get install make g++ libopencv-dev -y
-apt-get clean
-cd /usr/lib/x86_64-linux-gnu/pkgconfig/
-ln -sf opencv4.pc opencv.pc
-EOF
+RUN --mount=type=cache,target=/var/cache/apt \
+  apt-get update && \
+  apt-get install make g++ git libopencv-dev -y && \
+  apt-get clean && \
+  cd /usr/lib/x86_64-linux-gnu/pkgconfig/ && \
+  ln -sf opencv4.pc opencv.pc
 
-ARG BRANCH=main SHA=26e413ae9cf8dc04c617ca451a91a1624bfdf0c0
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-git fetch
-git reset --hard
-git checkout ${BRANCH}
-git reset --hard ${SHA}
-pip install -r binary_installer/py3.10-linux-x86_64-cuda-reqs.txt
-EOF
+
+
+ENV ROOT=/InvokeAI
+RUN git clone https://github.com/invoke-ai/InvokeAI.git ${ROOT}
+WORKDIR ${ROOT}
 
 RUN --mount=type=cache,target=/root/.cache/pip \
-  pip install -U --force-reinstall opencv-python-headless huggingface_hub && \
+  git reset --hard 4463124bddd221c333d4c70e73aa2949ad35453d && \
+  pip install .
+
+
+ARG BRANCH=main SHA=50eb02f68be912276a9c106d5e8038a5671a0386
+RUN --mount=type=cache,target=/root/.cache/pip \
+  git fetch && \
+  git reset --hard && \
+  git checkout ${BRANCH} && \
+  git reset --hard ${SHA} && \
+  pip install -U .
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+  --mount=type=bind,from=xformers,source=/wheel.whl,target=/xformers-0.0.17-cp310-cp310-linux_x86_64.whl \
+  pip install -U opencv-python-headless triton /xformers-0.0.17-cp310-cp310-linux_x86_64.whl && \
   python3 -c "from patchmatch import patch_match"
 
 
-RUN touch invokeai.init
 COPY . /docker/
 
-
-ENV ROOT=/stable-diffusion PYTHONPATH="${PYTHONPATH}:${ROOT}" PRELOAD=false CLI_ARGS=""
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV PYTHONUNBUFFERED=1 PRELOAD=false HF_HOME=/root/.cache/huggingface CONFIG_DIR=/data/config/invoke CLI_ARGS=""
 EXPOSE 7860
 
 ENTRYPOINT ["/docker/entrypoint.sh"]
-CMD python3 -u scripts/invoke.py --web --host 0.0.0.0 --port 7860 --config /docker/models.yaml --root_dir ${ROOT} --outdir /output/invoke ${CLI_ARGS}
+CMD invokeai --web --host 0.0.0.0 --port 7860 --root_dir ${ROOT} --config ${CONFIG_DIR}/models.yaml --outdir /output/invoke ${CLI_ARGS}
+# TODO: make sure the config is persisted between sessions

services/invoke/entrypoint.sh

@@ -4,25 +4,25 @@ set -Eeuo pipefail
 
 declare -A MOUNTS
 
+mkdir -p ${CONFIG_DIR}
+
 # cache
 MOUNTS["/root/.cache"]=/data/.cache/
 
+# this is really just a hack to avoid migrations
+rm -rf ${HF_HOME}/diffusers
+
 # ui specific
 MOUNTS["${ROOT}/models/codeformer"]=/data/Codeformer/
-
 MOUNTS["${ROOT}/models/gfpgan/GFPGANv1.4.pth"]=/data/GFPGAN/GFPGANv1.4.pth
-MOUNTS["${ROOT}/models/gfpgan/weights"]=/data/.cache/
-
+MOUNTS["${ROOT}/models/gfpgan/weights"]=/data/GFPGAN/
 MOUNTS["${ROOT}/models/realesrgan"]=/data/RealESRGAN/
 
-MOUNTS["${ROOT}/models/bert-base-uncased"]=/data/.cache/huggingface/transformers/
-MOUNTS["${ROOT}/models/openai/clip-vit-large-patch14"]=/data/.cache/huggingface/transformers/
-MOUNTS["${ROOT}/models/CompVis/stable-diffusion-safety-checker"]=/data/.cache/huggingface/transformers/
+MOUNTS["${ROOT}/models/ldm"]=/data/.cache/invoke/ldm/
 
 MOUNTS["${ROOT}/embeddings"]=/data/embeddings/
 
 # hacks
-MOUNTS["${ROOT}/models/clipseg"]=/data/.cache/invoke/clipseg/
 
 for to_path in "${!MOUNTS[@]}"; do
   set -Eeuo pipefail
@@ -40,7 +40,8 @@ done
 
 if "${PRELOAD}" == "true"; then
   set -Eeuo pipefail
-  python3 -u scripts/preload_models.py --skip-sd-weights --root ${ROOT} --config_file /docker/models.yaml
+  invokeai-configure --root ${ROOT} --yes
+  cp ${ROOT}/configs/models.yaml ${CONFIG_DIR}/models.yaml
 fi
 
 exec "$@"

services/invoke/models.yaml

@@ -1,23 +0,0 @@
-# This file describes the alternative machine learning models
-# available to InvokeAI script.
-#
-# To add a new model, follow the examples below. Each
-# model requires a model config file, a weights file,
-# and the width and height of the images it
-# was trained on.
-stable-diffusion-1.5:
-  description: Stable Diffusion version 1.5
-  weights: /data/StableDiffusion/v1-5-pruned-emaonly.ckpt
-  vae: /data/VAE/vae-ft-mse-840000-ema-pruned.ckpt
-  config: ./configs/stable-diffusion/v1-inference.yaml
-  width: 512
-  height: 512
-  default: true
-inpainting-1.5:
-  description: RunwayML SD 1.5 model optimized for inpainting
-  weights: /data/StableDiffusion/sd-v1-5-inpainting.ckpt
-  vae: /data/VAE/vae-ft-mse-840000-ema-pruned.ckpt
-  config: ./configs/stable-diffusion/v1-inpainting-inference.yaml
-  width: 512
-  height: 512
-  default: false

services/sygil/Dockerfile

@@ -1,45 +1,39 @@
-# syntax=docker/dockerfile:1
-
 FROM python:3.8-slim
 
-SHELL ["/bin/bash", "-ceuxo", "pipefail"]
-
 ENV DEBIAN_FRONTEND=noninteractive PIP_PREFER_BINARY=1
 
 RUN --mount=type=cache,target=/root/.cache/pip pip install torch==1.13.0 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117
 
 RUN apt-get update && apt install gcc libsndfile1 ffmpeg build-essential zip unzip git -y && apt-get clean
 
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-git config --global http.postBuffer 1048576000
-git clone https://github.com/Sygil-Dev/sygil-webui.git stable-diffusion
-cd stable-diffusion
-git reset --hard 5291437085bddd16d752f811b6552419a2044d12
-pip install -r requirements.txt
-EOF
+RUN --mount=type=cache,target=/root/.cache/pip \
+  git config --global http.postBuffer 1048576000 && \
+  git clone https://github.com/Sygil-Dev/sygil-webui.git stable-diffusion && \
+  cd stable-diffusion && \
+  git reset --hard 5291437085bddd16d752f811b6552419a2044d12 && \
+  pip install -r requirements.txt
 
 
 ARG BRANCH=master SHA=571fb897edd58b714bb385dfaa1ad59aecef8bc7
-RUN --mount=type=cache,target=/root/.cache/pip <<EOF
-cd stable-diffusion
-git fetch
-git checkout ${BRANCH}
-git reset --hard ${SHA}
-pip install -r requirements.txt
-EOF
+RUN --mount=type=cache,target=/root/.cache/pip \
+  cd stable-diffusion && \
+  git fetch && \
+  git checkout ${BRANCH} && \
+  git reset --hard ${SHA} && \
+  pip install -r requirements.txt
 
-RUN --mount=type=cache,target=/root/.cache/pip pip install transformers==4.24.0
+RUN --mount=type=cache,target=/root/.cache/pip pip install -U 'transformers>=4.24'
 
 # add info
 COPY . /docker/
-RUN <<EOF
-python /docker/info.py /stable-diffusion/frontend/frontend.py
-chmod +x /docker/mount.sh /docker/run.sh
-# streamlit
-sed -i -- 's/8501/7860/g' /stable-diffusion/.streamlit/config.toml
-EOF
+RUN python /docker/info.py /stable-diffusion/frontend/frontend.py && \
+  chmod +x /docker/mount.sh /docker/run.sh && \
+  # streamlit \
+  sed -i -- 's/8501/7860/g' /stable-diffusion/.streamlit/config.toml
 
 WORKDIR /stable-diffusion
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_VISIBLE_DEVICES=all
 ENV PYTHONPATH="${PYTHONPATH}:${PWD}" STREAMLIT_SERVER_HEADLESS=true USE_STREAMLIT=0 CLI_ARGS=""
 EXPOSE 7860