fix: fix manylinux2_28 build (#436)

guocuimi · web-flow · commit c85747f6e1a3 · 2025-04-23T15:08:34.000-07:00
diff --git a/.devcontainer/manylinux/Dockerfile b/.devcontainer/manylinux/Dockerfile
@@ -1,11 +1,11 @@
-ARG BASE_IMAGE=vectorchai/scalellm_manylinux:cuda12.4
+ARG BASE_IMAGE=vectorchai/scalellm_manylinux2_28:cuda12.6
 FROM ${BASE_IMAGE}
 
 ARG USER=vscode
 ARG UID=1000
 ARG GID=1000
-ARG TORCH_VERSION=2.6.0
-ARG CUDA_VERSION=12.4
+ARG TORCH_VERSION=2.7.0
+ARG CUDA_VERSION=12.6
 ARG PYTHON_VERSION=3.12
 
 # Run as non-root user
@@ -19,7 +19,7 @@ RUN yum -y update && yum -y install \
     htop \
     vim \
     jq \
-    gdb 
+    gdb
 
 # Choose python 3.12
 ENV PATH="/opt/python/cp312-cp312/bin:$PATH"
@@ -41,4 +41,4 @@ export CLICOLOR=1
 export LSCOLORS=ExFxCxDxBxegedabagacad
 EOT
 
-ENTRYPOINT [ "/usr/bin/bash" ]
+ENTRYPOINT [ "/usr/bin/bash" ]
diff --git a/.devcontainer/manylinux/devcontainer.json b/.devcontainer/manylinux/devcontainer.json
@@ -4,9 +4,9 @@
   "build": {
     "dockerfile": "Dockerfile",
     "args": {
-      "BASE_IMAGE": "vectorchai/scalellm_manylinux:cuda12.4",
-      "TORCH_VERSION": "2.6.0",
-      "CUDA_VERSION": "12.4",
+      "BASE_IMAGE": "vectorchai/scalellm_manylinux2_28:cuda12.6",
+      "TORCH_VERSION": "2.7.0",
+      "CUDA_VERSION": "12.6",
       "USER": "${localEnv:USER:vscode}",
       "UID": "${localEnv:UID:1000}",
       "GID": "${localEnv:GID:1000}"
@@ -66,4 +66,4 @@
       }
     }
   }
-}
+}
diff --git a/.github/workflows/build_wheel.yml b/.github/workflows/build_wheel.yml
@@ -43,11 +43,11 @@ jobs:
           mkdir -p $CI_CACHE_DIR/.vcpkg/bincache
           mkdir -p $CI_CACHE_DIR/.ccache
           mkdir -p $CI_CACHE_DIR/.pip
-      
+
       - name: Build wheel
         timeout-minutes: 60
         run: |
-          docker pull vectorchai/scalellm_manylinux:cuda${CUDA_VERSION}
+          docker pull vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION}
           docker run --rm -t \
               -v "$CI_CACHE_DIR":/ci_cache \
               -v "$GITHUB_WORKSPACE":/ScaleLLM \
@@ -60,7 +60,7 @@ jobs:
               -e CCACHE_DIR=/ci_cache/.ccache \
               -e PIP_CACHE_DIR=/ci_cache/.pip \
               -u $(id -u):$(id -g) \
-              vectorchai/scalellm_manylinux:cuda${CUDA_VERSION} \
+              vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION} \
               bash /ScaleLLM/scripts/build_wheel.sh
 
       - name: show wheel size
@@ -96,7 +96,7 @@ jobs:
           draft: false
           prerelease: false
           make_latest: true
-  
+
   commit_whl_index:
     needs: build_wheel
     runs-on: ubuntu-latest
@@ -110,7 +110,7 @@ jobs:
 
       - name: Show wheels
         run: ls -lha dist
-      
+
       - name: Clone whl repo
         run: |
           git clone https://oauth2:${WHL_TOKEN}@github.com/vectorch-ai/whl.git
@@ -130,5 +130,3 @@ jobs:
           git add -A
           git commit -m "Update whl index for version ${{ inputs.tag }}"
           git push
-      
-
diff --git a/.github/workflows/package_test.yml b/.github/workflows/package_test.yml
@@ -1,7 +1,7 @@
 name: Package test
 
 on:
-  workflow_dispatch: 
+  workflow_dispatch:
 
   push:
     paths:
@@ -11,7 +11,7 @@ on:
       - "requirements.txt"
       - "requirements-test.txt"
       - ".github/workflows/package_test.yml"
-    branches: 
+    branches:
       - main
 
   pull_request:
@@ -58,11 +58,11 @@ jobs:
         mkdir -p $CI_CACHE_DIR/.vcpkg/bincache
         mkdir -p $CI_CACHE_DIR/.ccache
         mkdir -p $CI_CACHE_DIR/.pip
-      
+
     - name: Build wheel
       timeout-minutes: 60
       run: |
-        docker pull vectorchai/scalellm_manylinux:cuda${CUDA_VERSION}
+        docker pull vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION}
         docker run --rm -t \
             -v "$CI_CACHE_DIR":/ci_cache \
             -v "$GITHUB_WORKSPACE":/ScaleLLM \
@@ -73,7 +73,7 @@ jobs:
             -e CCACHE_DIR=/ci_cache/.ccache \
             -e PIP_CACHE_DIR=/ci_cache/.pip \
             -u $(id -u):$(id -g) \
-            vectorchai/scalellm_manylinux:cuda${CUDA_VERSION} \
+            vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION} \
             bash /ScaleLLM/scripts/build_wheel.sh
 
     - name: Show whl package size
diff --git a/.github/workflows/publish_devel_image.yml b/.github/workflows/publish_devel_image.yml
@@ -10,7 +10,7 @@ on:
   # Schedule the workflow to run at 10:00 (UTC) every month.
   schedule:
     # Minute[0,59] Hour[0,23] Day of month[1,31] Month[1,12] Day of week[0,6] (Sunday=0)
-    - cron: '0 10 1 * *'  
+    - cron: '0 10 1 * *'
 
 env:
   # Tells where to store caches.
@@ -23,9 +23,6 @@ jobs:
       matrix:
         cuda: ["12.6", "12.8"]
         gcc: ["12"]
-        include: # build cuda 11.8 with gcc 11
-          - cuda: "11.8"
-            gcc: "11"
     runs-on: [self-hosted, linux, build]
     steps:
       - name: Checkout repository
@@ -61,4 +58,3 @@ jobs:
           tags: |
             vectorchai/scalellm_devel:cuda${{ matrix.cuda }}-ubuntu22.04
             vectorchai/scalellm_devel:cuda${{ matrix.cuda }}
-
diff --git a/.github/workflows/publish_manylinux_2_28_image.yml b/.github/workflows/publish_manylinux_2_28_image.yml
@@ -21,7 +21,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        cuda: ["12.6", "12.8"]
+        cuda: ["11.8", "12.6", "12.8"]
     runs-on: [self-hosted, linux, build]
     steps:
       - name: Checkout repository
diff --git a/.github/workflows/publish_wheel.yml b/.github/workflows/publish_wheel.yml
@@ -45,7 +45,7 @@ jobs:
       - name: Build wheel
         timeout-minutes: 60
         run: |
-          docker pull vectorchai/scalellm_manylinux:cuda${CUDA_VERSION}
+          docker pull vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION}
           docker run --rm -t \
               -v "$CI_CACHE_DIR":/ci_cache \
               -v "$GITHUB_WORKSPACE":/ScaleLLM \
@@ -57,16 +57,16 @@ jobs:
               -e CCACHE_DIR=/ci_cache/.ccache \
               -e PIP_CACHE_DIR=/ci_cache/.pip \
               -u $(id -u):$(id -g) \
-              vectorchai/scalellm_manylinux:cuda${CUDA_VERSION} \
+              vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION} \
               bash /ScaleLLM/scripts/build_wheel.sh
 
       - name: show wheels
         run: ls -lh dist
 
-      - name: rename wheel to manylinux
+      - name: rename wheel to manylinux_2_28
         run: |
           for whl in dist/scalellm-*.whl; do
-            new_whl=${whl//"-linux_"/"-manylinux1_"}
+            new_whl=${whl//"-linux_"/"-manylinux_2_28_"}
             if [ "$whl" != "$new_whl" ]; then
               mv $whl $new_whl
             fi
@@ -103,5 +103,3 @@ jobs:
           password: ${{ secrets.PYPI_API_TOKEN }}
           skip-existing: true
           verbose: true
-      
-
diff --git a/.github/workflows/release_test.yml b/.github/workflows/release_test.yml
@@ -8,7 +8,7 @@ on:
   # Schedule the workflow to run at 08:00 (UTC) every day.
   schedule:
     # Minute[0,59] Hour[0,23] Day of month[1,31] Month[1,12] Day of week[0,6] (Sunday=0)
-    - cron: '0 8 * * *' 
+    - cron: '0 8 * * *'
 
 env:
   # Tells where to store caches.
@@ -39,11 +39,11 @@ jobs:
         mkdir -p $CI_CACHE_DIR/.vcpkg/bincache
         mkdir -p $CI_CACHE_DIR/.ccache
         mkdir -p $CI_CACHE_DIR/.pip
-      
+
     - name: Build wheel
       timeout-minutes: 60
       run: |
-        docker pull vectorchai/scalellm_manylinux:cuda${CUDA_VERSION}
+        docker pull vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION}
         docker run --rm -t \
             -v "$CI_CACHE_DIR":/ci_cache \
             -v "$GITHUB_WORKSPACE":/ScaleLLM \
@@ -54,7 +54,7 @@ jobs:
             -e CCACHE_DIR=/ci_cache/.ccache \
             -e PIP_CACHE_DIR=/ci_cache/.pip \
             -u $(id -u):$(id -g) \
-            vectorchai/scalellm_manylinux:cuda${CUDA_VERSION} \
+            vectorchai/scalellm_manylinux2_28:cuda${CUDA_VERSION} \
             bash /ScaleLLM/scripts/build_wheel.sh
 
     - name: Show whl package size
diff --git a/docker/common/install_cmake.sh b/docker/common/install_cmake.sh
@@ -9,6 +9,7 @@ command -v pip3 >/dev/null && pip3 uninstall -y cmake
 
 # Remove existing CMake installation
 rm -f /usr/local/bin/cmake
+rm -f /opt/conda/bin/cmake
 
 path="v${CMAKE_VERSION}"
 ARCH=$(uname -m)
@@ -19,4 +20,4 @@ pushd /tmp
 wget -q "https://github.com/Kitware/CMake/releases/download/${path}/${file}"
 tar -C /usr/local --strip-components 1 --no-same-owner -zxf ${file}
 rm -f cmake-*.tar.gz
-popd
+popd

Original file line number	Diff line number	Diff line change
`@@ -4,9 +4,9 @@`
`4`	`4`	`"build": {`
`5`	`5`	`"dockerfile": "Dockerfile",`
`6`	`6`	`"args": {`
`7`		`- "BASE_IMAGE": "vectorchai/scalellm_manylinux:cuda12.4",`
`8`		`- "TORCH_VERSION": "2.6.0",`
`9`		`- "CUDA_VERSION": "12.4",`
	`7`	`+ "BASE_IMAGE": "vectorchai/scalellm_manylinux2_28:cuda12.6",`
	`8`	`+ "TORCH_VERSION": "2.7.0",`
	`9`	`+ "CUDA_VERSION": "12.6",`
`10`	`10`	`"USER": "${localEnv:USER:vscode}",`
`11`	`11`	`"UID": "${localEnv:UID:1000}",`
`12`	`12`	`"GID": "${localEnv:GID:1000}"`
`@@ -66,4 +66,4 @@`
`66`	`66`	`}`
`67`	`67`	`}`
`68`	`68`	`}`
`69`		`-}`
	`69`	`+}`