misc: upgrade cuda version and add devcontainer for manylinux (#412)

.devcontainer/devcontainer.json → .devcontainer/devel/devcontainer.json

@@ -1,6 +1,6 @@
 // For more config options, see https://aka.ms/devcontainer.json.
 {
-  "name": "cuda12.4",
+  "name": "devel:cu12.4",
   "build": {
     "dockerfile": "Dockerfile",
     "args": {

.devcontainer/manylinux/Dockerfile

@@ -0,0 +1,44 @@
+ARG BASE_IMAGE=vectorchai/scalellm_manylinux:cuda12.4
+FROM ${BASE_IMAGE}
+
+ARG USER=vscode
+ARG UID=1000
+ARG GID=1000
+ARG TORCH_VERSION=2.6.0
+ARG CUDA_VERSION=12.4
+ARG PYTHON_VERSION=3.12
+
+# Run as non-root user
+RUN groupadd --gid ${GID} ${USER} \
+    && useradd --uid ${UID} --gid ${GID} -m ${USER} --shell /bin/bash
+RUN echo ${USER} ALL=\(root\) NOPASSWD:ALL > /etc/sudoers.d/${USER} \
+    && chmod 0440 /etc/sudoers.d/${USER}
+
+# Install tools
+RUN yum -y update && yum -y install \
+    htop \
+    vim \
+    jq \
+    gdb 
+
+# Choose python 3.12
+ENV PATH="/opt/python/cp312-cp312/bin:$PATH"
+
+# Install torch and other dependencies
+RUN python -m pip install --upgrade pip
+RUN pip install --upgrade pip
+RUN pip install torch==$TORCH_VERSION -i "https://download.pytorch.org/whl/cu${CUDA_VERSION//./}"
+RUN pip install numpy jinja2
+RUN pip install --upgrade setuptools wheel
+
+# Set up bash prompt
+RUN cat <<EOT >> /home/${USER}/.bashrc
+parse_git_branch() {
+  git branch 2> /dev/null | sed -e '/^[^*]/d' -e 's/* \(.*\)/ (\1)/'
+}
+export PS1="\[\e[0;33m\]\u@\h \w\[\e[m\]\[\033[32m\]\$(parse_git_branch)\[\033[00m\]$ "
+export CLICOLOR=1
+export LSCOLORS=ExFxCxDxBxegedabagacad
+EOT
+
+ENTRYPOINT [ "/usr/bin/bash" ]

.devcontainer/manylinux/devcontainer.json

@@ -0,0 +1,69 @@
+// For more config options, see https://aka.ms/devcontainer.json.
+{
+  "name": "manylinux:cu12.4",
+  "build": {
+    "dockerfile": "Dockerfile",
+    "args": {
+      "BASE_IMAGE": "vectorchai/scalellm_manylinux:cuda12.4",
+      "TORCH_VERSION": "2.6.0",
+      "CUDA_VERSION": "12.4",
+      "USER": "${localEnv:USER:vscode}",
+      "UID": "${localEnv:UID:1000}",
+      "GID": "${localEnv:GID:1000}"
+    }
+  },
+  // Access GPUs from inside the container
+  "runArgs": [
+    "--gpus=all"
+  ],
+  // Environment variables passed to the container.
+  "containerEnv": {
+    "HUGGING_FACE_HUB_TOKEN": "${localEnv:HUGGING_FACE_HUB_TOKEN}"
+  },
+  // Run as the current user
+  "remoteUser": "${localEnv:USER:vscode}",
+  "containerUser": "${localEnv:USER:vscode}",
+  "updateRemoteUserUID": true,
+  // Ports should be forwarded from inside container to the local machine
+  "forwardPorts": [],
+  // For a full avaiable list, see https://containers.dev/features.
+  "features": {
+    "ghcr.io/devcontainers/features/git:1": {},
+    "ghcr.io/devcontainers/features/go:1": {}
+  },
+  "customizations": {
+    "vscode": {
+      "extensions": [
+        "ms-vscode.cmake-tools",
+        "ms-vscode.cpptools",
+        "ms-vscode.cpptools-extension-pack",
+        "llvm-vs-code-extensions.vscode-clangd",
+        "ms-python.python",
+        "ms-azuretools.vscode-docker",
+        "ziruiwang.nvidia-monitor",
+        "mutantdino.resourcemonitor"
+      ],
+      "settings": {
+        "extensions.verifySignature": false,
+        "terminal.integrated.defaultProfile.linux": "bash",
+        "C_Cpp.clang_format_sortIncludes": true,
+        "C_Cpp.default.intelliSenseMode": "linux-clang-x64",
+        "C_Cpp.default.cppStandard": "c++17",
+        "cmake.loggingLevel": "info",
+        "cmake.configureOnOpen": false,
+        "cmake.configureOnEdit": false,
+        "cmake.automaticReconfigure": true,
+        "cmake.ctest.allowParallelJobs": true,
+        "python.testing.pytestEnabled": true,
+        "python.analysis.inlayHints.pytestParameters": true,
+        "files.watcherExclude": {
+          "**/.git/objects/**": true,
+          "**/.git/subtree-cache/**": true,
+          "**/node_modules/*/**": true,
+          "third_party/**": true,
+          "build/**": true
+        }
+      }
+    }
+  }
+}

.github/workflows/build_wheel.yml

@@ -23,7 +23,7 @@ jobs:
     strategy:
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
-        cuda: ["11.8", "12.1", "12.4"]
+        cuda: ["11.8", "12.4", "12.6"]
         torch: ["2.4.1", "2.5.1", "2.6.0"]
     runs-on: [self-hosted, linux, release]
     env:

.github/workflows/publish_devel_image.yml

@@ -21,7 +21,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        cuda: ["12.1", "12.4", "12.6", "12.8"]
+        cuda: ["12.4", "12.6", "12.8"]
         gcc: ["12"]
         include: # build cuda 11.8 with gcc 11
           - cuda: "11.8"

.github/workflows/publish_manylinux_image.yml

@@ -21,17 +21,21 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        cuda: ["11.8", "12.1", "12.4", "12.6"]
+        cuda: ["11.8", "12.4", "12.6"]
     runs-on: [self-hosted, linux, build]
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3
+        with:
+          cache-image: false
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
+        with:
+          cache-binary: false
 
       - name: Login to Docker Hub
         uses: docker/login-action@v3

docker/Dockerfile.devel

@@ -27,7 +27,7 @@ RUN if [ -n "${CMAKE_VERSION}" ]; then bash ./install_cmake.sh; fi
 RUN rm install_cmake.sh
 RUN cmake --version
 
-ARG NINJA_VERSION=1.11.1
+ARG NINJA_VERSION=1.12.1
 COPY ./common/install_ninja.sh install_ninja.sh
 RUN if [ -n "${NINJA_VERSION}" ]; then bash ./install_ninja.sh; fi
 RUN rm install_ninja.sh

docker/Dockerfile.manylinux

@@ -8,7 +8,7 @@ ENV LC_ALL=en_US.UTF-8
 ENV LANG=en_US.UTF-8
 ENV LANGUAGE=en_US.UTF-8
 
-# Install dependencies for vcpkg
+# Install dependencies
 RUN yum -y update && yum -y install \
     zip \
     unzip \
@@ -17,7 +17,9 @@ RUN yum -y update && yum -y install \
     curl \
     perl \
     perl-IPC-Cmd \
-    sudo 
+    sudo \
+    automake \
+    libtool
 
 # Show versions of gcc, g++, nvcc
 RUN gcc --version; g++ --version; nvcc --version

docker/Dockerfile.manylinux_2_28

@@ -30,7 +30,7 @@ ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:$PATH
 # RUN if [ -n "${CMAKE_VERSION}" ]; then bash ./install_cmake.sh; fi
 # RUN rm install_cmake.sh
 
-ARG NINJA_VERSION=1.11.1
+ARG NINJA_VERSION=1.12.1
 COPY ./common/install_ninja.sh install_ninja.sh
 RUN if [ -n "${NINJA_VERSION}" ]; then bash ./install_ninja.sh; fi
 RUN rm install_ninja.sh

docker/common/install_base.sh

@@ -21,7 +21,8 @@ install_ubuntu() {
     jq \
     libtool \
     unzip \
-    gdb
+    gdb \
+    automake
 
   # Cleanup package manager
   apt-get autoclean && apt-get clean
@@ -39,7 +40,8 @@ install_almalinux() {
     vim \
     jq \
     libtool \
-    unzip
+    unzip \
+    automake
 
   # Cleanup
   yum clean all

docker/common/install_cmake.sh

@@ -11,7 +11,8 @@ command -v pip3 >/dev/null && pip3 uninstall -y cmake
 rm -f /usr/local/bin/cmake
 
 path="v${CMAKE_VERSION}"
-file="cmake-${CMAKE_VERSION}-linux-x86_64.tar.gz"
+ARCH=$(uname -m)
+file="cmake-${CMAKE_VERSION}-linux-${ARCH}.tar.gz"
 
 # Download and install specific CMake version in /usr/local
 pushd /tmp