ci: use cuda 12.9 for devel and 12.8 for test (#471)

guocuimi · web-flow · commit 674e388df927 · 2025-06-18T22:06:41.000-07:00
diff --git a/.devcontainer/devel/Dockerfile b/.devcontainer/devel/Dockerfile
@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=vectorchai/scalellm_devel:cuda12.4
+ARG BASE_IMAGE=vectorchai/scalellm_devel:cuda12.9
 FROM ${BASE_IMAGE}
 
 ARG USER=vscode
@@ -21,4 +21,4 @@ export CLICOLOR=1
 export LSCOLORS=ExFxCxDxBxegedabagacad
 EOT
 
-ENTRYPOINT [ "/usr/bin/bash" ]
+ENTRYPOINT [ "/usr/bin/bash" ]
diff --git a/.devcontainer/devel/devcontainer.json b/.devcontainer/devel/devcontainer.json
@@ -1,10 +1,10 @@
 // For more config options, see https://aka.ms/devcontainer.json.
 {
-  "name": "devel:cu12.4",
+  "name": "devel:cu12.9",
   "build": {
     "dockerfile": "Dockerfile",
     "args": {
-      "BASE_IMAGE": "vectorchai/scalellm_devel:cuda12.4",
+      "BASE_IMAGE": "vectorchai/scalellm_devel:cuda12.9",
       "USER": "${localEnv:USER:vscode}",
       "UID": "${localEnv:UID:1000}",
       "GID": "${localEnv:GID:1000}"
@@ -64,4 +64,4 @@
       }
     }
   }
-}
+}
diff --git a/.devcontainer/manylinux/Dockerfile b/.devcontainer/manylinux/Dockerfile
@@ -4,8 +4,8 @@ FROM ${BASE_IMAGE}
 ARG USER=vscode
 ARG UID=1000
 ARG GID=1000
-ARG TORCH_VERSION=2.7.0
-ARG CUDA_VERSION=12.6
+ARG TORCH_VERSION=2.7.1
+ARG CUDA_VERSION=12.8
 ARG PYTHON_VERSION=3.12
 
 # Run as non-root user
diff --git a/.devcontainer/manylinux/devcontainer.json b/.devcontainer/manylinux/devcontainer.json
@@ -1,12 +1,12 @@
 // For more config options, see https://aka.ms/devcontainer.json.
 {
-  "name": "manylinux:cu12.4",
+  "name": "manylinux:cu12.8",
   "build": {
     "dockerfile": "Dockerfile",
     "args": {
-      "BASE_IMAGE": "vectorchai/scalellm_manylinux2_28:cuda12.6",
-      "TORCH_VERSION": "2.7.0",
-      "CUDA_VERSION": "12.6",
+      "BASE_IMAGE": "vectorchai/scalellm_manylinux2_28:cuda12.8",
+      "TORCH_VERSION": "2.7.1",
+      "CUDA_VERSION": "12.8",
       "USER": "${localEnv:USER:vscode}",
       "UID": "${localEnv:UID:1000}",
       "GID": "${localEnv:GID:1000}"
diff --git a/.github/workflows/create_release.yml b/.github/workflows/create_release.yml
@@ -25,7 +25,7 @@ jobs:
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
         cuda: ["11.8", "12.6", "12.8"]
-        torch: ["2.7.0"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
diff --git a/.github/workflows/pkg_build.yml b/.github/workflows/pkg_build.yml
@@ -40,7 +40,7 @@ jobs:
       matrix:
         python: ["3.12"]
         cuda: ["12.8"]
-        torch: ["2.7.0"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
@@ -92,7 +92,7 @@ jobs:
       matrix:
         python: ["3.12"]
         cuda: ["12.8"]
-        torch: ["2.7.0"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux, gpu]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
diff --git a/.github/workflows/pkg_test.yml b/.github/workflows/pkg_test.yml
@@ -20,8 +20,8 @@ jobs:
       fail-fast: false
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
-        cuda: ["12.6"]
-        torch: ["2.7.0"]
+        cuda: ["12.8"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
@@ -73,8 +73,8 @@ jobs:
       fail-fast: false
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
-        cuda: ["12.6"]
-        torch: ["2.7.0"]
+        cuda: ["12.8"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux, gpu]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
diff --git a/.github/workflows/publish_pypi.yml b/.github/workflows/publish_pypi.yml
@@ -23,7 +23,7 @@ jobs:
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
         cuda: ["12.6"]
-        torch: ["2.7.0"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
@@ -89,7 +89,7 @@ jobs:
       matrix:
         python: ["3.9", "3.10", "3.11", "3.12"]
         cuda: ["12.6"]
-        torch: ["2.7.0"]
+        torch: ["2.7.1"]
     runs-on: [self-hosted, linux, gpu]
     env:
       PYTHON_VERSION: ${{ matrix.python }}
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -175,14 +175,14 @@ if (DEFINED ENV{LIBTORCH_ROOT})
 else()
   include(FetchContent)
   if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL 12.8)
-    # download libtorch 2.7.0 with cuda 12.8 from pytorch.org
-    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu128/libtorch-cxx11-abi-shared-with-deps-2.7.0%2Bcu128.zip")
+    # download libtorch 2.7.1 with cuda 12.8 from pytorch.org
+    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu128/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu128.zip")
   elseif(CUDAToolkit_VERSION VERSION_GREATER_EQUAL 12.6)
-    # download libtorch 2.7.0 with cuda 12.6 from pytorch.org
-    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu126/libtorch-cxx11-abi-shared-with-deps-2.7.0%2Bcu126.zip")
+    # download libtorch 2.7.1 with cuda 12.6 from pytorch.org
+    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu126/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu126.zip")
   elseif(CUDAToolkit_VERSION VERSION_GREATER_EQUAL 11.8)
-    # download libtorch 2.7.0 with cuda 11.8 from pytorch.org
-    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu118/libtorch-cxx11-abi-shared-with-deps-2.7.0%2Bcu118.zip")
+    # download libtorch 2.7.1 with cuda 11.8 from pytorch.org
+    set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cu118/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu118.zip")
   else()
     # error out if cuda version is not supported
     message(FATAL_ERROR "Unsupported CUDA version: ${CUDAToolkit_VERSION}")
@@ -201,7 +201,7 @@ else()
   FetchContent_MakeAvailable(libtorch)
 
   find_package(Torch REQUIRED PATHS ${libtorch_SOURCE_DIR} NO_DEFAULT_PATH)
-  message(STATUS "Downloading and using libtorch 2.7.0 for cuda ${CUDA_VERSION} at ${libtorch_SOURCE_DIR}")
+  message(STATUS "Downloading and using libtorch 2.7.1 for cuda ${CUDA_VERSION} at ${libtorch_SOURCE_DIR}")
 endif()
 
 # carry over torch flags to the rest of the project
diff --git a/docs/source/index.rst b/docs/source/index.rst
@@ -12,7 +12,7 @@ ScaleLLM is available as a Python Wheel package on `PyPI <https://pypi.org/proje
 
 .. code-block:: bash
 
-   # Install ScaleLLM with CUDA 12.6 and Pytorch 2.7.0
+   # Install ScaleLLM with CUDA 12.6 and Pytorch 2.7.1
    $ pip install -U scalellm
 
 
diff --git a/docs/source/quick_start.rst b/docs/source/quick_start.rst
@@ -25,31 +25,31 @@ If you want to install ScaleLLM with different versions of CUDA and PyTorch, you
 
         .. tabs::
 
-            .. tab:: PyTorch 2.7.0
+            .. tab:: PyTorch 2.7.1
 
                 .. code-block:: bash
 
-                    $ pip install -U scalellm -i https://whl.vectorch.com/cu128/torch2.7.0/
+                    $ pip install -U scalellm -i https://whl.vectorch.com/cu128/torch2.7.1/
 
     .. tab:: CUDA 12.6
 
         .. tabs::
 
-            .. tab:: PyTorch 2.7.0
+            .. tab:: PyTorch 2.7.1
 
                 .. code-block:: bash
 
-                    $ pip install -U scalellm -i https://whl.vectorch.com/cu126/torch2.7.0/
+                    $ pip install -U scalellm -i https://whl.vectorch.com/cu126/torch2.7.1/
 
     .. tab:: CUDA 11.8
 
         .. tabs::
 
-            .. tab:: PyTorch 2.7.0
+            .. tab:: PyTorch 2.7.1
 
                 .. code-block:: bash
 
-                    $ pip install -U scalellm -i https://whl.vectorch.com/cu118/torch2.7.0/
+                    $ pip install -U scalellm -i https://whl.vectorch.com/cu118/torch2.7.1/
 
 
 Build from source
@@ -65,7 +65,7 @@ If no wheel package is available for your configuration, you can build ScaleLLM
 
 
 Inference
--------
+---------
 
 You can use ScaleLLM for offline batch inference or online distributed inference.
 
diff --git a/setup.py b/setup.py
@@ -170,16 +170,21 @@ def build_extension(self, ext: CMakeExtension):
         debug = int(os.environ.get("DEBUG", 0)) if self.debug is None else self.debug
         build_type = "Debug" if debug else "Release"
 
-        cuda_version = get_cuda_version()
-        cuda_architectures = ["80", "89"]
-        if cuda_version >= Version("12.8"):
-            # blackwell needs cuda 12.8
-            cuda_architectures += ["90a", "100a", "120a"]
-        elif cuda_version >= Version("12.1"):
-            # hopper needs cuda 12.1
-            cuda_architectures += ["90a"]
+        if self.inplace:
+            # use native for develop
+            cuda_architectures = ["native"]
         else:
-            cuda_architectures += ["90"]
+            # decide cuda architectures based on cuda version
+            cuda_version = get_cuda_version()
+            cuda_architectures = ["80", "89"]
+            if cuda_version >= Version("12.8"):
+                # blackwell needs cuda 12.8
+                cuda_architectures += ["90a", "100a", "120a"]
+            elif cuda_version >= Version("12.1"):
+                # hopper needs cuda 12.1
+                cuda_architectures += ["90a"]
+            else:
+                cuda_architectures += ["90"]
 
         cmake_args = [
             "-G",
diff --git a/tools/run_in_docker.sh b/tools/run_in_docker.sh
@@ -38,7 +38,7 @@ function usage() {
 (( $# < 1 )) && usage
 
 # Default image
-IMAGE="vectorchai/scalellm_devel:cuda12.8"
+IMAGE="vectorchai/scalellm_devel:cuda12.9"
 RUN_OPTS=("--gpus=all")
 
 INTERACTIVE=1

Original file line number	Diff line number	Diff line change
`@@ -1,10 +1,10 @@`
`1`	`1`	`// For more config options, see https://aka.ms/devcontainer.json.`
`2`	`2`	`{`
`3`		`- "name": "devel:cu12.4",`
	`3`	`+ "name": "devel:cu12.9",`
`4`	`4`	`"build": {`
`5`	`5`	`"dockerfile": "Dockerfile",`
`6`	`6`	`"args": {`
`7`		`- "BASE_IMAGE": "vectorchai/scalellm_devel:cuda12.4",`
	`7`	`+ "BASE_IMAGE": "vectorchai/scalellm_devel:cuda12.9",`
`8`	`8`	`"USER": "${localEnv:USER:vscode}",`
`9`	`9`	`"UID": "${localEnv:UID:1000}",`
`10`	`10`	`"GID": "${localEnv:GID:1000}"`
`@@ -64,4 +64,4 @@`
`64`	`64`	`}`
`65`	`65`	`}`
`66`	`66`	`}`
`67`		`-}`
	`67`	`+}`