Merge pull request #33 from aisingapore/0.3.1-cvgpu

[Fix] 0.3.1 - Get GPU image to use GPU resources

.gitlab-ci.yml

@@ -10,12 +10,13 @@ pages:
   before_script:
     - pip install -r mkdocs-requirements.txt
   script:
-    - mkdocs build --strict --verbose --site-dir public
+    - mkdocs build --verbose --site-dir public
   artifacts:
     paths:
       - public
   rules:
     - if: $CI_COMMIT_REF_NAME == $CI_DEFAULT_BRANCH
+  needs: []
 
 update:onprem-runai:
   stage: update
@@ -40,4 +41,5 @@ update:onprem-runai:
     - cp -rv ../git-repo/.git .
     - git add .
     - git commit -m "$CI_COMMIT_MESSAGE" || echo "Nothing to commit"
-    - git push origin $CI_COMMIT_BRANCH || echo "Nothing to push"
+    - git push origin $CI_COMMIT_BRANCH || echo "Nothing to push"
+  needs: []

hooks/post_gen_project.py

@@ -4,21 +4,9 @@
 
 def populate_problem(problem_domain: str) -> None:
 
-    SUB_DIRS = ["src", "conf", "notebooks", "aisg-context"]
-
-    for subdir in SUB_DIRS:
-
-        working_dir = os.getcwd()
-        src_dir = os.path.join(working_dir, "problem-templates", problem_domain)
-        shutil.copytree(
-            os.path.join(src_dir, subdir),
-            os.path.join(working_dir, subdir),
-            dirs_exist_ok=True
-        )
-        shutil.copy2(
-            os.path.join(src_dir, "{{cookiecutter.repo_name}}-conda-env.yaml"),
-            os.path.join(working_dir, "{{cookiecutter.repo_name}}-conda-env.yaml")
-        )
+    working_dir = os.getcwd()
+    src_dir = os.path.join(working_dir, "problem-templates", problem_domain)
+    shutil.copytree(src_dir, working_dir, dirs_exist_ok=True)
 
 
 def generate_template_scripts() -> None:

{{cookiecutter.repo_name}}/aisg-context/guide-site/docs/guide-for-user/07-job-orchestration.md

@@ -312,6 +312,16 @@ and connect to http://localhost:5000.
 We shall build the Docker image from the Docker file 
 `docker/{{cookiecutter.repo_name}}-gpu.Dockerfile`:
 
+!!! warning "Attention"
+
+    If you're only using CPUs for training, then you can just use
+    `docker/{{cookiecutter.repo_name}}-cpu.Dockerfile` instead for
+    smaller image size.  
+    If you're using AMD GPUs for training, you can copy the components
+    from the [`rocm`][rocm] folder in the Kapitan Hull repository.
+
+[rocm]: https://github.com/aisingapore/kapitan-hull/tree/main/rocm
+
 === "Linux/macOS"
 
     ```bash

{{cookiecutter.repo_name}}/problem-templates/cv/aisg-context/guide-site/docs/guide-for-user/07-job-orchestration.md

@@ -326,6 +326,16 @@ and connect to http://localhost:5000.
 We shall build the Docker image from the Docker file 
 `docker/{{cookiecutter.repo_name}}-gpu.Dockerfile`:
 
+!!! warning "Attention"
+
+    If you're only using CPUs for training, then you can just use
+    `docker/{{cookiecutter.repo_name}}-cpu.Dockerfile` instead for
+    smaller image size.  
+    If you're using AMD GPUs for training, you can copy the components
+    from the [`rocm`][rocm] folder in the Kapitan Hull repository.
+
+[rocm]: https://github.com/aisingapore/kapitan-hull/tree/main/rocm
+
 === "Linux/macOS"
 
     ```bash

{{cookiecutter.repo_name}}/problem-templates/cv/docker/{{cookiecutter.repo_name}}-gpu.Dockerfile

@@ -0,0 +1,53 @@
+FROM ubuntu:20.04
+
+ARG DEBIAN_FRONTEND="noninteractive"
+
+ARG NON_ROOT_USER="aisg"
+ARG NON_ROOT_UID="2222"
+ARG NON_ROOT_GID="2222"
+ARG HOME_DIR="/home/${NON_ROOT_USER}"
+
+ARG REPO_DIR="."
+ARG CONDA_ENV_FILE="{{cookiecutter.repo_name}}-conda-env-gpu.yaml"
+ARG CONDA_ENV_NAME="{{cookiecutter.repo_name}}"
+
+# Miniconda arguments
+ARG CONDA_HOME="/miniconda3"
+ARG CONDA_BIN="${CONDA_HOME}/bin/conda"
+ARG CONDA_VER="py310_23.5.2-0"
+ARG CONDA_ARCH="Linux-x86_64"
+ARG MINICONDA_SH="Miniconda3-${CONDA_VER}-${CONDA_ARCH}.sh"
+
+RUN useradd -l -m -s /bin/bash -u ${NON_ROOT_UID} ${NON_ROOT_USER} && \
+    mkdir -p ${CONDA_HOME} && \
+    chown -R ${NON_ROOT_USER}:${NON_ROOT_GID} ${CONDA_HOME}
+
+RUN apt-get update && \
+    apt-get -y install bzip2 curl wget gcc rsync git vim locales && \
+    sed -i -e 's/# en_US.UTF-8 UTF-8/en_US.UTF-8 UTF-8/' /etc/locale.gen && \
+    locale-gen && \
+    dpkg-reconfigure --frontend=noninteractive locales && \
+    update-locale LANG=en_US.UTF-8 && \
+    apt-get clean
+
+ENV PYTHONIOENCODING utf8
+ENV LANG "C.UTF-8"
+ENV LC_ALL "C.UTF-8"
+
+USER ${NON_ROOT_USER}
+WORKDIR ${HOME_DIR}
+
+COPY --chown=${NON_ROOT_USER}:${NON_ROOT_GID} ${REPO_DIR} {{cookiecutter.repo_name}}
+
+# Install Miniconda
+RUN curl -O https://repo.anaconda.com/miniconda/$MINICONDA_SH && \
+    chmod +x ${MINICONDA_SH} && \
+    ./${MINICONDA_SH} -u -b -p ${CONDA_HOME} && \
+    rm ${MINICONDA_SH}
+ENV PATH ${CONDA_HOME}/bin:${HOME_DIR}/.local/bin:$PATH
+
+# Install conda environment
+RUN ${CONDA_BIN} env create -f {{cookiecutter.repo_name}}/${CONDA_ENV_FILE} && \
+    ${CONDA_BIN} init bash && \
+    ${CONDA_BIN} clean -a -y && \
+    echo "source activate ${CONDA_ENV_NAME}" >> "${HOME_DIR}/.bashrc"

{{cookiecutter.repo_name}}/problem-templates/cv/{{cookiecutter.repo_name}}-conda-env-gpu.yaml

@@ -0,0 +1,26 @@
+name: {{cookiecutter.repo_name}}
+channels:
+  - defaults
+  - pytorch
+  - nvidia
+  - conda-forge
+dependencies:
+  - pytorch=2.1.2
+  - torchvision=0.16.2
+  - pytorch-cuda=12.1
+  - python=3.11.7
+  - pip=23.3.2
+  - pip:
+    - mlflow-skinny==2.9.2
+    - hydra-core==1.3.2
+    - hydra-optuna-sweeper==1.2.0
+    - python-json-logger==2.0.7
+    - fastapi==0.109.0
+    - uvicorn[standard]==0.25.0
+    - python-multipart==0.0.6
+    - jsonlines==4.0.0
+    - pandas==2.1.4
+    - gunicorn==21.2.0
+    - pydantic==2.5.3
+    - pydantic-settings==2.1.0
+    - ipykernel==6.25.0