diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 9b40fb36c303fdd53fe94b0ab874f6b77e66c20d..e3431f20792505ae0f975e90aef7734516625daf 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -180,3 +180,26 @@ test_keras3cpu.image:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
changes:
- keras3-cpu/Dockerfile
+
+pytorchgpu.image:
+ stage: build
+ variables:
+ REGISTRY_IMAGE: christofkaufmann/pytorch-gpu-notebook
+ script:
+ - echo "{\"auths\":{\"https://index.docker.io/v1/\":{\"auth\":\"$(printf "%s:%s" "${REGISTRY_USER}" "${REGISTRY_PASSWORD}" | base64 | tr -d '\n')\"}}}" > /kaniko/.docker/config.json
+ - /kaniko/executor
+ --context "dir://${CI_PROJECT_DIR}/pytorch-gpu"
+ --dockerfile "${CI_PROJECT_DIR}/pytorch-gpu/Dockerfile"
+ --target build
+ --build-arg LABEL_CREATED=$(date -u +'%Y-%m-%dT%H:%M:%SZ')
+ --build-arg LABEL_REVISION=${CI_COMMIT_SHA}
+ --destination "${REGISTRY_IMAGE}:${CI_COMMIT_SHORT_SHA}"
+ --destination "${REGISTRY_IMAGE}"
+ --cache=false
+ --compressed-caching=false
+ --use-new-run
+ --snapshot-mode=redo
+ rules:
+ - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
+ changes:
+ - pytorch-gpu/Dockerfile
diff --git a/README.md b/README.md
index d73460e0853a6cb5e6a3087488670409d93f9d1c..34ec9b5527c6d08c0a44db2956c066baa2e3798d 100644
--- a/README.md
+++ b/README.md
@@ -47,6 +47,18 @@ This is a collection of customized Jupyter images to be used in the JupyterHub r
- Added Notebook git Puller ([nbgitpuller](https://hub.jupyter.org/nbgitpuller))
- Provide default home directory contents including VS Code Extensions and settings in `/etc/skel/`. Can be copied on first startup using [post start hook](https://z2jh.jupyter.org/en/stable/jupyterhub/customizing/user-environment.html#about-user-storage-and-adding-files-to-it).
+## pytorch-gpu
+
+- Based on: `jupyter/scipy-notebook:latest` ([quay.io](https://quay.io/repository/jupyter/scipy-notebook), [src](https://github.com/jupyter/docker-stacks/tree/main/images/scipy-notebook), [doc](https://jupyter-docker-stacks.readthedocs.io/en/latest/using/selecting.html#jupyter-scipy-notebook))
+- Available as: [`christofkaufmann/pytorch-gpu-notebook:latest`](https://hub.docker.com/repository/docker/christofkaufmann/pytorch-gpu-notebook)
+- Features:
+ - Jupyter enabled
+ - Contains PyTorch with GPU support (requires nvidia-driver)
+ - Contains Huggingface packages
+ - Added VS Code via code-server as web UI and via VS Code CLI for tunnels
+ - Added Notebook git Puller ([nbgitpuller](https://hub.jupyter.org/nbgitpuller))
+ - Provide default home directory contents including VS Code Extensions and settings in `/etc/skel/`. Can be copied on first startup using [post start hook](https://z2jh.jupyter.org/en/stable/jupyterhub/customizing/user-environment.html#about-user-storage-and-adding-files-to-it).
+
## Usage
You can use the image from docker hub or build it locally.
@@ -103,7 +115,7 @@ docker exec -it tf-cpu bash
docker stop tf-cpu # also removes container, because of the --rm parameter on docker run
```
-And the same for the GPU images (`tensorflow-gpu-with-vscode` and `keras3-gpu`):
+And the same for the GPU images (`tensorflow-gpu-with-vscode`, `keras3-gpu`, `pytorch-gpu`):
```bash
cd tensorflow-gpu-with-vscode
diff --git a/pytorch-gpu/Dockerfile b/pytorch-gpu/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..86914565d1646c04103c2e18301fc4d3cc2ee87b
--- /dev/null
+++ b/pytorch-gpu/Dockerfile
@@ -0,0 +1,268 @@
+# syntax=docker/dockerfile:1
+
+# use scipy-notebook and install torch from PyPI after conda-forge packages
+FROM quay.io/jupyter/scipy-notebook:latest AS build
+
+# fix: https://github.com/hadolint/hadolint/wiki/DL4006
+# fix: https://github.com/koalaman/shellcheck/wiki/SC3014
+SHELL ["/bin/bash", "-o", "pipefail", "-c"]
+
+USER root
+
+# install code-server and extensions
+ENV CODE_VERSION=4.93.1
+RUN wget --no-hsts -q https://github.com/coder/code-server/releases/download/v$CODE_VERSION/code-server_${CODE_VERSION}_amd64.deb && \
+ dpkg -i code-server_${CODE_VERSION}_amd64.deb && \
+ rm -f code-server_${CODE_VERSION}_amd64.deb && \
+ code-server --force \
+ --install-extension ms-python.python \
+ --install-extension ms-toolsai.jupyter \
+ --install-extension eamodio.gitlens \
+ --install-extension gitlab.gitlab-workflow && \
+ mkdir -p /usr/local/bin/start-notebook.d && \
+ chown -R ${NB_USER} "/home/${NB_USER}/.config" "/home/${NB_USER}/.local" && \
+ fix-permissions "/home/${NB_USER}"
+
+# install VS Code CLI
+RUN wget --no-hsts -q -O vscode_cli.tar.gz 'https://code.visualstudio.com/sha/download?build=stable&os=cli-alpine-x64' && \
+ tar -xf vscode_cli.tar.gz --directory /usr/local/bin && \
+ rm vscode_cli.tar.gz && \
+ chown ${NB_USER} /usr/local/bin/code && \
+ fix-permissions /usr/local/bin/code
+
+# install some packages
+# NOTE: without --no-update-deps somehow an undefined symbol error in /opt/conda/lib/python3.11/lib-dynload/_sqlite3.cpython-311-x86_64-linux-gnu.so occurs, which crashes jupyter lab.
+USER ${NB_UID}
+RUN mamba install --yes --no-update-deps \
+ # umap + hdbscan
+ 'umap-learn' \
+ 'hdbscan' \
+ # gradio
+ 'gradio' \
+ # optuna + plotly
+ 'optuna' \
+ 'plotly' \
+ # optional dependencies for pandas for read_html and to support Parquet files
+ 'lxml' \
+ 'pyarrow' \
+ 'pyogrio' \
+ # geopandas
+ 'geopandas' \
+ 'pysal' \
+ # networkx + pyvis + netgraph
+ 'networkx' \
+ 'pyvis' \
+ 'netgraph' \
+ # dependencies for jupyter-vscode-proxy (?)
+ 'rfc3339-validator' \
+ 'rfc3986-validator' \
+ 'uri-template' \
+ 'fqdn' \
+ 'webcolors' \
+ 'isoduration' \
+ 'jsonpointer' \
+ # provide a way for lecturers to share code
+ 'nbgitpuller' \
+ # monitor GPUs in terminal
+ 'nvtop' \
+ # integration of VS Code in JupyterLab
+ 'jupyter-server-proxy' \
+ 'jupyter-vscode-proxy' \
+ # some improvements of jupyterlab
+ 'jupyterlab_execute_time' \
+ 'jupyter-archive' \
+ 'jupyter-resource-usage' \
+ # install opencv via fastai, headless version not available in conda-forge
+ 'fastai::opencv-python-headless' && \
+ pip install --no-cache-dir --extra-index-url 'https://pypi.nvidia.com' --extra-index-url 'https://download.pytorch.org/whl/cu118' \
+ 'transformers' \
+ 'diffusers' \
+ 'datasets' \
+ 'timm' \
+ 'torch' \
+ 'torchaudio' \
+ 'torchvision' && \
+ printf '%s\n' \
+ "" \
+ "# track CPU usage, prevent known bug with prometheus, set default limits if not set by MEM_LIMIT and CPU_LIMIT" \
+ "import os" \
+ "" \
+ "c.ResourceUseDisplay.track_cpu_percent = True" \
+ "c.ResourceUseDisplay.enable_prometheus_metrics = False" \
+ "c.ResourceUseDisplay.mem_limit = int(os.getenv('MEM_LIMIT', os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES')))" \
+ "c.ResourceUseDisplay.cpu_limit = float(os.getenv('CPU_LIMIT', os.cpu_count()))" \
+ >> /etc/jupyter/jupyter_notebook_config.py && \
+ mamba clean --all -f -y && \
+ fix-permissions "${CONDA_DIR}" && \
+ fix-permissions "/home/${NB_USER}"
+
+# default settings and extensions in /etc/skel
+USER root
+# activate extensions by default
+COPY plugin.jupyterlab-settings "/home/${NB_USER}/.jupyter/lab/user-settings/@jupyterlab/extensionmanager-extension/"
+RUN chown -R ${NB_USER} "/home/${NB_USER}/.jupyter" && \
+ # history search with Page Up/Down
+ sed -i "s/^# \(.*history-search.*\)/\1/" /etc/inputrc && \
+ # VS Code Python settings
+ mkdir -p "/home/${NB_USER}/.local/share/code-server/User" && \
+ echo -e '{\n "python.defaultInterpreterPath": "/opt/conda/bin/python",\n "jupyter.sendSelectionToInteractiveWindow": true\n}' > /home/${NB_USER}/.local/share/code-server/User/settings.json && \
+ mkdir -p "/home/${NB_USER}/.local/share/code-server/Machine" && \
+ cp "/home/${NB_USER}/.local/share/code-server/User/settings.json" "/home/${NB_USER}/.local/share/code-server/Machine/settings.json" && \
+ mkdir -p /etc/skel/.local/share/ && \
+ cp -r "/home/${NB_USER}/.local/share/code-server" /etc/skel/.local/share/ && \
+ # matplotlib cache
+ mkdir -p /etc/skel/.cache && \
+ cp -r "/home/${NB_USER}/.cache/matplotlib" /etc/skel/.cache/ && \
+ # stuff
+ cp -r "/home/${NB_USER}/.conda" "/home/${NB_USER}/.config" "/home/${NB_USER}/.jupyter" /etc/skel/ && \
+ # VS Code extension "gitlab-workflow", default gitlab server: gitlab.cvh-server.de
+ printf '%s\n' \
+ '# Set default gitlab server to gitlab.cvh-server.de' \
+ 'relpath=".local/share/code-server/extensions/gitlab.gitlab-workflow-*/extension.js"' \
+ 'for abspath in /home/${NB_USER}/${relpath} /etc/skel/${relpath}; do' \
+ ' if [[ -e "${abspath}" ]]; then' \
+ ' sed -i "s_=\"https://gitlab.com\"_=\"https://gitlab.cvh-server.de\"_g" "${abspath}"' \
+ ' fi' \
+ 'done' \
+ >> /usr/local/bin/start-notebook.d/fix-gitlab-server.sh && \
+ # source mamba.sh in bash using /etc/profile.d/conda.sh
+ mamba init --system && \
+ chown -R ${NB_USER} "/home/${NB_USER}/.local" && \
+ fix-permissions "/home/${NB_USER}" && \
+ fix-permissions "/etc/skel"
+# remember to copy back in post start hook
+
+
+# override maintainer label from Jupyter docker stacks
+LABEL maintainer="Christof Kaufmann <christof.kaufmann@hs-bochum.de>"
+# OCI annotations, see https://github.com/opencontainers/image-spec/blob/main/annotations.md
+ARG LABEL_CREATED
+ARG LABEL_REVISION=test-build
+LABEL org.opencontainers.image.created=$LABEL_CREATED
+LABEL org.opencontainers.image.authors="Christof Kaufmann <christof.kaufmann@hs-bochum.de>"
+LABEL org.opencontainers.image.source="https://gitlab.cvh-server.de/ckaufmann/gpu-cluster-images"
+LABEL org.opencontainers.image.revision=$LABEL_REVISION
+LABEL org.opencontainers.image.vendor="UAS Bochum"
+LABEL org.opencontainers.image.licenses=BSD-3-Clause
+LABEL org.opencontainers.image.title="Jupyter Notebook PyTorch GPU image"
+LABEL org.opencontainers.image.description="This image includes PyTorch with GPU support, Huggingface, VS Code CLI, code-server and nbgitpuller."
+LABEL org.opencontainers.image.base.name=quay.io/jupyter/scipy-notebook:latest
+
+# switch back to jovyan to avoid accidental container runs as root
+USER ${NB_UID}
+WORKDIR "${HOME}"
+
+
+###################################################################
+######################## Testing the image ########################
+###################################################################
+FROM build AS test
+
+# replace home directory with skel, which is closer to the kubernetes environment
+USER root
+RUN cd / && \
+ rm -r "/home/${NB_USER}" && \
+ cp -r /etc/skel "/home/${NB_USER}" && \
+ fix-permissions "/home/${NB_USER}"
+USER ${NB_UID}
+
+# enable logging for all Jupyter applications
+RUN printf '%s\n' \
+ "" \
+ "# Filter out user connection hint message, which is a CRITICAL logger message" \
+ "import logging" \
+ "class UserHintFilter(logging.Filter):" \
+ " def filter(self, record):" \
+ " return 'To access the server' not in record.getMessage()" \
+ "" \
+ "c.Application.logging_config = {" \
+ " 'filters': {" \
+ " 'user_hint': {" \
+ " '()': UserHintFilter," \
+ " }," \
+ " }," \
+ " 'formatters': {" \
+ " 'file': {" \
+ " 'format': '%(asctime)s %(levelname)-8s %(name)-15s %(message)s'," \
+ " }," \
+ " }," \
+ " 'handlers': {" \
+ " 'file': {" \
+ " 'class': 'logging.FileHandler'," \
+ " 'filters': ['user_hint']," \
+ " 'formatter': 'file'," \
+ " 'level': 'INFO'," \
+ " 'filename': '/home/jovyan/jupyter.log'," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'Application': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'JupyterApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'ExtensionApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'LabServerApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'LabApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'NotebookApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ " 'loggers': {" \
+ " 'ServerApp': {" \
+ " 'level': 'DEBUG'," \
+ " 'handlers': ['console', 'file']," \
+ " }," \
+ " }," \
+ "}" \
+ >> /etc/jupyter/jupyter_notebook_config.py
+
+
+# collect lab logs for 4 sec
+RUN start-notebook.sh & sleep 4 && kill -INT %1 && sleep 1
+
+# check log file:
+ # not existing means some error prevents logging → bad
+RUN [[ -f jupyter.log ]] \
+ || ( echo "Log file jupyter.log does not exist!" && false ) \
+ # existing, contains error → bad, print log file
+ && ! grep -qP '(Traceback|ERROR|CRITICAL)' jupyter.log \
+ || ( cat jupyter.log && false )
+
+
+# add some environment variables and collect lab logs again for 4 sec
+RUN mv jupyter.log jupyter1.log
+ENV MEM_LIMIT=1000000000 \
+ CPU_LIMIT=16.0
+RUN start-notebook.sh & sleep 4 && kill -INT %1 && sleep 1
+
+# check log file:
+ # not existing means some error prevents logging → bad
+RUN [[ -f jupyter.log ]] \
+ || ( echo "Log file jupyter.log does not exist!" && false ) \
+ # existing, contains error → bad, print log file
+ && ! grep -qP '(Traceback|ERROR|CRITICAL)' jupyter.log \
+ || ( cat jupyter.log && false )
diff --git a/pytorch-gpu/plugin.jupyterlab-settings b/pytorch-gpu/plugin.jupyterlab-settings
new file mode 100644
index 0000000000000000000000000000000000000000..43938cf61322cf74f501e9d6e42858d979557095
--- /dev/null
+++ b/pytorch-gpu/plugin.jupyterlab-settings
@@ -0,0 +1,15 @@
+{
+ // Extension Manager
+ // @jupyterlab/extensionmanager-extension:plugin
+ // Extension manager settings.
+ // *********************************************
+
+ // Disclaimed Status
+ // Whether the user agrees the access to external web services and understands extensions may introduce security risks or contain malicious code that runs on his machine.
+ "disclaimed": true,
+
+ // Enabled Status
+ // Enables extension manager.
+ // WARNING: installing untrusted extensions may be unsafe.
+ "enabled": true
+}
\ No newline at end of file