[fix] added manylinux support (#185)

- [x] fix the build error for cuda 11.8
- [ ] add cuda 11.8 support for manylinux


/usr/local/cuda/bin/../targets/x86_64-linux/include/crt/host_config.h:132:2:
error: #error -- unsupported GNU version! gcc versions later than 11 are
not supported! The nvcc flag '-allow-unsupported-compiler' can be used
to override this version check; however, using an unsupported host
compiler may cause compilation failure or incorrect run time execution.
Use at your own risk.
132 | #error -- unsupported GNU version! gcc versions later than 11 are
not supported! The nvcc flag '-allow-unsupported-compiler' can be used
to override this version check; however, using an unsupported host
compiler may cause compilation failure or incorrect run time execution.
Use at your own risk.

.github/workflows/publish_devel_image.yml

@@ -1,6 +1,9 @@
 name: Publish devel docker image
 on:
   workflow_dispatch:
+env:
+  # Tells where to store caches.
+  CI_CACHE_DIR: ${{ github.workspace }}/../../ci_cache
 
 jobs:
   publish_base:
@@ -27,6 +30,8 @@ jobs:
           context: ./docker
           file: ./docker/Dockerfile.devel
           push: true
+          cache-from: type=local,src=$CI_CACHE_DIR/.buildx-cache
+          cache-to: type=local,dest=$CI_CACHE_DIR/.buildx-cache
           build-args: |
             UBUNTU_VERSION=22.04
             CUDA_VERSION=12.1

.github/workflows/publish_manylinux_image.yml

@@ -0,0 +1,52 @@
+name: Publish manylinux docker image
+on:
+  workflow_dispatch:
+env:
+  # Tells where to store caches.
+  CI_CACHE_DIR: ${{ github.workspace }}/../../ci_cache
+
+jobs:
+  publish_base:
+    runs-on: [self-hosted, linux, release]
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_HUB_USER }}
+          password: ${{ secrets.DOCKER_HUB_TOKEN }}
+
+      - name: Build base for cuda 12.1
+        uses: docker/build-push-action@v5
+        with:
+          context: ./docker
+          file: ./docker/Dockerfile.manylinux
+          push: true
+          cache-from: type=local,src=$CI_CACHE_DIR/.buildx-cache
+          cache-to: type=local,dest=$CI_CACHE_DIR/.buildx-cache
+          build-args: |
+            CUDA_VERSION=12.1
+          tags: |
+            vectorchai/scalellm_manylinux:cuda12.1
+
+      # - name: Build base for cuda 11.8
+      #   uses: docker/build-push-action@v5
+      #   with:
+      #     context: ./docker
+      #     file: ./docker/Dockerfile.manylinux
+      #     push: true
+      #     cache-from: type=local,src=$CI_CACHE_DIR/.buildx-cache
+      #     cache-to: type=local,dest=$CI_CACHE_DIR/.buildx-cache
+      #     build-args: |
+      #       CUDA_VERSION=11.8
+      #     tags: |
+      #       vectorchai/scalellm_manylinux:cuda11.8
+

.github/workflows/release_wheel.yml

@@ -13,10 +13,11 @@ env:
 jobs:
   build_wheel:
     strategy:
+      fail-fast: false
       matrix:
-        python: ["3.9", "3.10", "3.11", "3.12"]
-        cuda: ["11.8", "12.1"]
-        torch: ["2.1", "2.2", "2.3"]
+        python: ["3.9", "3.10", "3.11"]
+        cuda: ["12.1"]
+        torch: ["2.2", "2.3"]
     runs-on: [self-hosted, linux, release]
     steps:
       - name: Checkout repository
@@ -36,7 +37,7 @@ jobs:
               -e VCPKG_DEFAULT_BINARY_CACHE=/ci_cache/.vcpkg/bincache \
               -e CCACHE_DIR=/ci_cache/.ccache \
               --user $(id -u):$(id -g) \
-              vectorchai/scalellm_builder:cuda${{ matrix.cuda }}-ubuntu22.04 \
+              vectorchai/scalellm_manylinux:cuda${{ matrix.cuda }} \
               bash /ScaleLLM/scripts/build_wheel.sh
         timeout-minutes: 60
 

CMakeLists.txt

@@ -3,6 +3,7 @@ set_property(GLOBAL PROPERTY USE_FOLDERS ON)
 
 option(USE_CCACHE "Attempt using CCache to wrap the compilation" ON)
 option(USE_CXX11_ABI "Use the new C++-11 ABI, which is not backwards compatible." ON)
+option(USE_MANYLINUX "Build for manylinux" OFF)
 
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
@@ -144,7 +145,6 @@ if(UNIX)
   set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -Og")
 endif()
 
-find_package(Boost CONFIG REQUIRED)
 find_package(Threads REQUIRED)
 # find all dependencies from vcpkg
 find_package(fmt CONFIG REQUIRED)
@@ -162,7 +162,13 @@ find_package(prometheus-cpp CONFIG REQUIRED)
 find_package(stduuid CONFIG REQUIRED)
 find_package(RapidJSON CONFIG REQUIRED)
 
-find_package(Python REQUIRED COMPONENTS Interpreter Development)
+if (USE_MANYLINUX)
+  # manylinux doesn't ship Development.Embed
+  find_package(Python REQUIRED COMPONENTS Interpreter Development.Module)
+else()
+  find_package(Python REQUIRED COMPONENTS Interpreter Development)
+endif()
+
 find_package(NCCL REQUIRED)
 
 if (USE_CXX11_ABI)

docker/Dockerfile.base

@@ -29,7 +29,7 @@ ENV DESIRED_CUDA ${CUDA_VERSION}
 ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:$PATH
 
 # Install gcc
-ARG GCC_VERSION=12
+ARG GCC_VERSION=11
 RUN apt-get update \
     && apt-get install -y --no-install-recommends \
     software-properties-common gpg-agent

docker/Dockerfile.manylinux

@@ -0,0 +1,48 @@
+FROM quay.io/pypa/manylinux_2_28_x86_64 as base
+
+LABEL maintainer="mi@vectorch.com"
+ENV DEBIAN_FRONTEND noninteractive
+
+ENV LC_ALL en_US.UTF-8
+ENV LANG en_US.UTF-8
+ENV LANGUAGE en_US.UTF-8
+
+# Install common dependencies
+COPY ./common/install_base.sh install_base.sh
+RUN bash ./install_base.sh && rm install_base.sh
+
+# Install user
+COPY ./common/install_user.sh install_user.sh
+RUN bash ./install_user.sh && rm install_user.sh
+
+# Install cuda, cudnn and nccl
+ARG CUDA_VERSION=12.1
+COPY ./common/install_cuda.sh install_cuda.sh
+RUN bash ./install_cuda.sh ${CUDA_VERSION} && rm install_cuda.sh
+ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:$PATH
+
+# ARG CMAKE_VERSION=3.18.5
+# COPY ./common/install_cmake.sh install_cmake.sh
+# RUN if [ -n "${CMAKE_VERSION}" ]; then bash ./install_cmake.sh; fi
+# RUN rm install_cmake.sh
+
+ARG NINJA_VERSION=1.11.1
+COPY ./common/install_ninja.sh install_ninja.sh
+RUN if [ -n "${NINJA_VERSION}" ]; then bash ./install_ninja.sh; fi
+RUN rm install_ninja.sh
+
+ARG CCACHE_VERSION=4.8.3
+COPY ./common/install_ccache.sh install_ccache.sh
+RUN if [ -n "${CCACHE_VERSION}" ]; then bash ./install_ccache.sh; fi
+RUN rm install_ccache.sh
+
+# install rust
+ENV RUSTUP_HOME=/usr/local/rustup
+ENV CARGO_HOME=/usr/local/cargo
+ENV PATH=/usr/local/cargo/bin:$PATH
+RUN curl https://sh.rustup.rs -sSf | sh -s -- -y
+# give everyone permission to use rust
+RUN chmod -R a+w ${RUSTUP_HOME} ${CARGO_HOME}
+RUN rustup --version; cargo --version; rustc --version
+
+CMD ["bash"]

docker/common/install_base.sh

@@ -31,11 +31,31 @@ install_ubuntu() {
   rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
 }
 
+install_almalinux() {
+  yum -y update
+  yum -y install \
+    zip \
+    wget \
+    curl \
+    perl \
+    sudo \
+    vim \
+    jq \
+    libtool \
+    unzip
+
+  # Cleanup
+  yum clean all
+}
+
 ID=$(grep -oP '(?<=^ID=).+' /etc/os-release | tr -d '"')
 case "$ID" in
   ubuntu)
     install_ubuntu
     ;;
+  almalinux)
+    install_almalinux
+    ;;
   *)
     echo "Unable to determine OS..."
     exit 1

docker/common/install_ccache.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+
+set -ex
+
+[ -n "$CCACHE_VERSION" ]
+
+ARCH=$(uname -m)
+url=https://github.com/ccache/ccache/releases/download/v${CCACHE_VERSION}/ccache-${CCACHE_VERSION}-linux-${ARCH}.tar.xz
+
+pushd /tmp
+curl -L "$url" | xz -d | tar -x
+cp ./ccache-${CCACHE_VERSION}-linux-x86_64/ccache /usr/bin/ccache
+popd
+
+# set max cache size to 5GiB
+/usr/bin/ccache -M 5Gi

docker/common/install_cmake.sh

@@ -10,6 +10,9 @@ case "$ID" in
   ubuntu)
     apt-get remove cmake -y
     ;;
+  almalinux)
+    rm /usr/local/bin/cmake
+    ;;
   *)
     echo "Unable to determine OS..."
     exit 1