[UPDATE] Remove conda; Use UV! Torch is also removed

jamesnulliu · jamesnulliu · commit e985a3d04f8e · 2025-08-22T14:50:53.000-07:00
diff --git a/.github/workflows/cd-docker-build-push.yml b/.github/workflows/cd-docker-build-push.yml
@@ -15,10 +15,6 @@ jobs:
           "Dockerfile.cuda", 
           "Dockerfile.cpu"
         ]
-        install_torch: [
-          true, 
-          false
-        ]
 
     steps:
       - name: Free Disk Space (Ubuntu)
@@ -45,8 +41,6 @@ jobs:
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: Build Docker image
-        env:
-          INSTALL_TORCH: ${{ matrix.install_torch }}
         run: |
           IMAGE_NAME=$(bash ./scripts/build.sh "${{ matrix.dockerfile }}")
           echo "IMAGE_NAME=$IMAGE_NAME" >> $GITHUB_ENV
diff --git a/Dockerfile.cpu b/Dockerfile.cpu
@@ -10,22 +10,23 @@ ENV LANG=en_US.UTF-8
 ARG INSTALL_TORCH
 ARG LLVM_VERSION
 ARG IMAGE_VERSION
-ARG TORCH_VERSION
+ARG IMAGE_NAME
 
 LABEL maintainer="JamesNULLiu jamesnulliu@gmail.com"
 LABEL version=${IMAGE_VERSION}
 
+ENV IMAGE_NAME=${IMAGE_NAME}
 ENV VCPKG_ROOT=/opt/vcpkg
 ENV VCPKG_HOME=/opt/vcpkg
-ENV CONDA_HOME=/opt/miniconda3
+ENV UV_HOME=/opt/uv
 ENV ENV_SETUP_FILE=/etc/local/env_setup.sh
 
 SHELL ["/bin/bash", "-c"]
 
 # Some basic tools
 RUN apt-get update && apt-get upgrade -y && \
     apt-get install -y \
-    apt-utils lsb-release software-properties-common gnupg git \
+    apt-utils lsb-release software-properties-common gnupg git acl sed\
     vim-gtk3 wget p7zip-full ninja-build curl jq pkg-config ssh \
     build-essential gdb htop tmux ttf-mscorefonts-installer kmod \
     libssl-dev && \
@@ -65,35 +66,17 @@ RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
 # Config files
 COPY data/vimrc /etc/vim/vimrc
 COPY data/env_setup.sh ${ENV_SETUP_FILE}
-# [TODO] The config files should not only be copied only to /root, but also to
-#        /etc/skel, so that the new user can also use them. But in singularity,
-#        /etc/skel has no effect, so I will decide what to do later.
 COPY data/.inputrc data/.bashrc data/.tmux.conf /root/
 COPY data/.inputrc data/.bashrc data/.tmux.conf /etc/skel/
+COPY data/create-shared-dir /usr/local/bin/
 
-# Install Miniconda3 and conda env
-# [TODO] Conda now uses python=3.13 in default. However, some packages (i.e.,
-#        vllm) only support python<=3.12. That's why I install python=3.12 
-#        mannually here. Maybe some days later I will remove this command.
-RUN wget -O /tmp/miniconda3.sh \
-    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
-    mkdir -p ${CONDA_HOME} && \
-    bash /tmp/miniconda3.sh -b -u -p ${CONDA_HOME} && \
-    ${CONDA_HOME}/bin/conda tos accept && \
-    \. ${CONDA_HOME}/bin/activate && \
-    conda upgrade libstdcxx-ng -c conda-forge -y && \
-    conda install -y python=3.12 && \
-    pip3 install nvitop --no-cache-dir && \
-    if [[ "${INSTALL_TORCH}" == "true" ]]; then \
-    pip3 install torch==${TORCH_VERSION} torchvision torchaudio \
-    --index-url https://download.pytorch.org/whl/cpu \
-    --no-cache-dir \
-    ; fi
+# Install uv
+RUN curl -LsSf https://astral.sh/uv/install.sh | \
+        env UV_INSTALL_DIR="${UV_HOME}" UV_NO_MODIFY_PATH=1 sh
 
 # Some final steps
 RUN apt-get update && apt-get upgrade -y && apt-get autoremove -y && \
     apt-get clean && rm -rf /var/lib/apt/lists/* && \
-    conda clean --all -y && pip cache purge && \
     git config --system --unset-all user.name  || true && \
     git config --system --unset-all user.email || true && \
     git config --global --unset-all user.name  || true && \
diff --git a/Dockerfile.cuda b/Dockerfile.cuda
@@ -7,28 +7,28 @@ ARG DEBIAN_FRONTEND=noninteractive
 ENV LANGUAGE=en_US.UTF-8
 ENV LANG=en_US.UTF-8
 
-ARG INSTALL_TORCH
 ARG LLVM_VERSION
 ARG IMAGE_VERSION
-ARG TORCH_VERSION
+ARG IMAGE_NAME
 
 LABEL maintainer="JamesNULLiu jamesnulliu@gmail.com"
 LABEL version=${IMAGE_VERSION}
 
+ENV IMAGE_NAME=${IMAGE_NAME}
 ENV CUDA_HOME=/usr/local/cuda
 ENV VCPKG_ROOT=/opt/vcpkg
 ENV VCPKG_HOME=/opt/vcpkg
-ENV CONDA_HOME=/opt/miniconda3
+ENV UV_HOME=/opt/uv
 ENV ENV_SETUP_FILE=/etc/local/env_setup.sh
 
 SHELL ["/bin/bash", "-c"]
 
 # Some basic tools
 RUN apt-get update && apt-get upgrade -y && \
     apt-get install -y \
-    apt-utils lsb-release software-properties-common gnupg git \
+    apt-utils lsb-release software-properties-common gnupg git acl sed \
     vim-gtk3 wget p7zip-full ninja-build curl jq pkg-config ssh \
-    build-essential gdb htop tmux ttf-mscorefonts-installer  kmod \
+    build-essential gdb htop tmux ttf-mscorefonts-installer kmod \
     libssl-dev && \
     fc-cache -f -v
 
@@ -66,36 +66,17 @@ RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
 # Config files
 COPY data/vimrc /etc/vim/vimrc
 COPY data/env_setup.sh ${ENV_SETUP_FILE}
-# [TODO] The config files should not only be copied only to /root, but also to
-#        /etc/skel, so that the new user can also use them. But in singularity,
-#        /etc/skel has no effect, so I will decide what to do later.
 COPY data/.inputrc data/.bashrc data/.tmux.conf /root/
 COPY data/.inputrc data/.bashrc data/.tmux.conf /etc/skel/
+COPY data/create-shared-dir /usr/local/bin/
 
-# Install Miniconda3 and conda env
-# [TODO] Conda now uses python=3.13 in default. However, some packages (i.e.,
-#        vllm) only support python<=3.12. That's why I install python=3.12 
-#        mannually here. Maybe some days later I will remove this command.
-RUN wget -O /tmp/miniconda3.sh \
-    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
-    mkdir -p ${CONDA_HOME} && \
-    bash /tmp/miniconda3.sh -b -u -p ${CONDA_HOME} && \
-    ${CONDA_HOME}/bin/conda tos accept && \
-    \. ${CONDA_HOME}/bin/activate && \
-    conda upgrade libstdcxx-ng -c conda-forge -y && \
-    conda install -y python=3.12 && \
-    pip3 install nvitop --no-cache-dir && \
-    if [[ "${INSTALL_TORCH}" == "true" ]]; then \
-    TORCH_CU_VER=$(echo $CUDA_VERSION | cut -d'.' -f1,2 | tr -d '.') && \
-    pip3 install torch==${TORCH_VERSION} torchvision torchaudio \
-    --index-url "https://download.pytorch.org/whl/cu${TORCH_CU_VER}" \
-    --no-cache-dir \
-    ; fi
+# Install uv
+RUN curl -LsSf https://astral.sh/uv/install.sh | \
+        env UV_INSTALL_DIR="${UV_HOME}" UV_NO_MODIFY_PATH=1 sh
 
 # Some final steps
 RUN apt-get update && apt-get upgrade -y && apt-get autoremove -y && \
     apt-get clean && rm -rf /var/lib/apt/lists/* && \
-    conda clean --all -y && pip cache purge && \
     git config --system --unset-all user.name  || true && \
     git config --system --unset-all user.email || true && \
     git config --global --unset-all user.name  || true && \
diff --git a/data/create-shared-dir b/data/create-shared-dir
@@ -0,0 +1,63 @@
+#!/bin/bash
+
+set -e
+
+# ==============================================================================
+# Script: create-shared-dir
+# Description: Configures one or more directories to be shared with a specific group.
+#
+# It performs the following actions on each target directory:
+#   1. Creates the directory if it doesn't exist.
+#   2. Recursively sets the group ownership.
+#   3. Sets permissions (group: rwx, others: none).
+#   4. Sets the 'setgid' bit on subdirectories to enforce group inheritance.
+#   5. Uses Access Control Lists (ACLs) to enforce permissions for new items.
+#
+# Usage:
+#   sudo create-shared-dir <shared_group> <dir1> [<dir2> ...]
+#
+# Example:
+#   sudo create-shared-dir developers /srv/data /home/shared/project
+#
+# ==============================================================================
+
+# --- Argument Parsing ---
+SHARED_GROUP="$1"
+shift # Shift arguments so $@ contains only the directories
+TARGET_DIRS=("$@")
+
+# --- Input Validation ---
+if [ ${#TARGET_DIRS[@]} -eq 0 ] || [ "$1" == "-h" ] || [ "$1" == "--help" ]; then
+    echo "Usage: $0 <shared_group> <dir1> [<dir2> ...]"
+    echo "Example: $0 developers /var/www/project_a"
+    exit 1
+fi
+
+# --- Pre-flight Checks ---
+# 1. Check for root privileges.
+if [[ $EUID -ne 0 ]]; then
+    echo "Error: This script must be run as root (or with sudo)." >&2
+    exit 1
+fi
+
+# 2. Check if the specified group exists.
+if ! getent group "$SHARED_GROUP" > /dev/null; then
+    echo "Error: Group '$SHARED_GROUP' does not exist." >&2
+    exit 1
+fi
+
+# --- Main Logic ---
+echo "Configuring shared directories for group '$SHARED_GROUP'..."
+
+for target_dir in "${TARGET_DIRS[@]}"; do
+    echo "--> Processing: $target_dir"
+    mkdir -p "$target_dir"
+    chgrp -R "$SHARED_GROUP" "$target_dir"
+    chmod -R g=rwX,o-rwx "$target_dir"
+    find "$target_dir" -type d -exec chmod g+s {} +
+    setfacl -R -m "g:$SHARED_GROUP:rwX" "$target_dir"
+    setfacl -R -d -m "g:$SHARED_GROUP:rwX" "$target_dir"
+done
+
+echo "Configuration complete."
+exit 0
diff --git a/data/env_setup.sh b/data/env_setup.sh
@@ -1,4 +1,23 @@
-echo "[ENV-SETUP] Setting up environment with: ${ENV_SETUP_FILE}"
+set -e
+
+sed "s|\${IMAGE_NAME}|${IMAGE_NAME}|g" << 'EOF'
+===============================================================================
+                 <-. (`-')_                ,-.  _(`-')              
+                    \( OO) )   <-.         / / ( (OO ).->    <-.    
+          <-.--. ,--./ ,--/  ,--. )       / /   \    .'_   ,--. )   
+        (`-'| ,| |   \ |  |  |  (`-')    / /    '`'-..__)  |  (`-') 
+        (OO |(_| |  . '|  |) |  |OO )   / /     |  |  ' |  |  |OO ) 
+       ,--. |  | |  |\    | (|  '__ |  / /      |  |  / : (|  '__ | 
+       |  '-'  / |  | \   |  |     |' / /       |  '-'  /  |     |' 
+        `-----'  `--'  `--'  `-----'  `-'       `------'   `-----'  
+-------------------------------------------------------------------------------
+Image Name:    ${IMAGE_NAME}
+Creator:       jamesnulliu <jamesnulliu@gmail.com>
+License:       MIT
+===============================================================================
+EOF
+
+echo "[ENV-SETUP] Sourcing \${ENV_SETUP_FILE}: ${ENV_SETUP_FILE}"
 
 # @brief Add `$1` into environment variable `$2` if it is not already there.
 # @example > env_load PATH /usr/local/bin
@@ -25,31 +44,6 @@ env_unload() {
     export $var_name=$(IFS=:; echo "${new_paths[*]}")
 }
 
-if [ -d "${CONDA_HOME:-}" ]; then
-    if [[ $- == *i* ]]; then
-        # Initialize conda in interactive mode
-        __conda_setup="$('${CONDA_HOME}/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
-        if [ $? -eq 0 ]; then
-            eval "$__conda_setup"
-        else
-            if [ -f "${CONDA_HOME}/etc/profile.d/conda.sh" ]; then
-                . "${CONDA_HOME}/etc/profile.d/conda.sh"
-            else
-                env_load PATH "${CONDA_HOME}/bin"
-            fi
-        fi
-        unset __conda_setup
-        echo "[ENV-SETUP] Conda initialized in interactive mode from $CONDA_HOME"
-    else
-        # Initialize conda in non-interactive mode
-        # Note that you may have to use `conda run` to use conda environments
-        . "${CONDA_HOME}/etc/profile.d/conda.sh" 
-        echo "[ENV-SETUP] Conda initialized in non-interactive mode from $CONDA_HOME."
-    fi
-else
-    unset CONDA_HOME
-fi
-
 if [ -d "${CUDA_HOME:-}" ]; then
     alias LOAD_CUDA="env_load PATH $CUDA_HOME/bin && \
         env_load LD_LIBRARY_PATH $CUDA_HOME/lib64"
@@ -71,4 +65,18 @@ if [ -d "${VCPKG_HOME:-}" ]; then
     echo "[ENV-SETUP] VCPKG initialized from $VCPKG_HOME"
 else
     unset VCPKG_HOME
+fi
+
+if [ -d "${UV_HOME:-}" ]; then
+    alias LOAD_UV="env_load PATH $UV_HOME"
+    alias UNLOAD_UV="env_unload PATH $UV_HOME"
+    env_load PATH "$UV_HOME"
+    echo "[ENV-SETUP] UV initialized from $UV_HOME"
+    echo "|- Note: For managing shared UV cache directory:"
+    echo "|-   1. Export UV_CACHE_DIR to a target directory. (No need to create it first.)"
+    echo "|-   2. Run: \`sudo create-shared-dir <group> \${UV_CACHE_DIR}\`" 
+    echo "|-   3. For all users in <group>, make sure UV_CACHE_DIR is set properly."
+    echo "|-      You can add it to \${ENV_SETUP_FILE} for convenience."
+else
+    unset UV_HOME
 fi
diff --git a/scripts/build.sh b/scripts/build.sh
@@ -1,3 +1,5 @@
+#!/bin/bash
+
 set -e 
 
 DOCKER_FILE=$1
@@ -10,20 +12,14 @@ elif [[ "${DOCKER_FILE}" == *.cuda ]]; then
     IMAGE_TAG="${IMAGE_TAG}-cuda${CUDA_VERSION}"
 fi
 
-if [[ "${INSTALL_TORCH}" == "true" ]]; then 
-    IMAGE_TAG="${IMAGE_TAG}-torch${TORCH_VERSION}"
-fi
-
 IMAGE_NAME=jamesnulliu/deeplearning:${IMAGE_TAG}
 
 docker build \
     -f $DOCKER_FILE \
-    --build-arg INSTALL_TORCH=$INSTALL_TORCH \
     --build-arg IMAGE_VERSION=$IMAGE_VERSION \
-    --build-arg TORCH_VERSION=$TORCH_VERSION \
+    --build-arg IMAGE_NAME=$IMAGE_NAME \
     --build-arg CUDA_VERSION=$CUDA_VERSION \
     --build-arg UBUNTU_VERSION=$UBUNTU_VERSION \
-    --build-arg CUDNN_VERSION=$CUDNN_VERSION \
     --build-arg LLVM_VERSION=$LLVM_VERSION \
     -t $IMAGE_NAME \
     .
diff --git a/scripts/exec.sh b/scripts/exec.sh
@@ -1,3 +1,5 @@
+#!/bin/bash
+
 CONTAINER_NAME=$1
 
 if [ ! $CONTAINER_NAME ]; then
diff --git a/scripts/image-configs.sh b/scripts/image-configs.sh
@@ -1,4 +1,4 @@
-IMAGE_VERSION=2.3.0
+IMAGE_VERSION=2.4.0
 TORCH_VERSION=2.7.1
 CUDA_VERSION=12.8.0
 UBUNTU_VERSION=24.04
diff --git a/scripts/run.sh b/scripts/run.sh
@@ -1,3 +1,5 @@
+#!/bin/bash
+
 set -e
 
 IMAGE_NAME=$1
@@ -12,6 +14,7 @@ if [ ! $CONTAINER_NAME ]; then
     CONTAINER_NAME=tmp
 fi
 
+if [[ "$IMAGE_NAME" == *"cuda"* ]]; then
 docker run -td \
     --name $CONTAINER_NAME \
     --gpus all \
@@ -20,6 +23,15 @@ docker run -td \
     --hostname $CONTAINER_NAME \
     -v $HOME/Projects:/root/Projects \
     $IMAGE_NAME
+else
+docker run -td \
+    --name $CONTAINER_NAME \
+    --network=host \
+    --shm-size=16G \
+    --hostname $CONTAINER_NAME \
+    -v $HOME/Projects:/root/Projects \
+    $IMAGE_NAME
+fi
 
 docker cp $HOME/.ssh $CONTAINER_NAME:/root/
 docker start $CONTAINER_NAME

Original file line number	Diff line number	Diff line change
`@@ -1,3 +1,5 @@`
	`1`	`+#!/bin/bash`
	`2`	`+`
`1`	`3`	`CONTAINER_NAME=$1`
`2`	`4`
`3`	`5`	`if [ ! $CONTAINER_NAME ]; then`
Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-IMAGE_VERSION=2.3.0`
	`1`	`+IMAGE_VERSION=2.4.0`
`2`	`2`	`TORCH_VERSION=2.7.1`
`3`	`3`	`CUDA_VERSION=12.8.0`
`4`	`4`	`UBUNTU_VERSION=24.04`