【Hackathon 9th No.109】[CppExtension] 添加 fastdeploy_ops 目录到 package_data 以支持现代打包方式 (#5156)

---------

Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com>
Co-authored-by: SigureMo <sigure.qaq@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

.github/workflows/_accuracy_test.yml

@@ -143,7 +143,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_base_test.yml

@@ -143,7 +143,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_build_linux.yml

@@ -155,7 +155,7 @@ jobs:
             elif [[ "${PADDLEVERSION}" != "" ]];then
               python -m pip install paddlepaddle-gpu==${PADDLEVERSION} -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
             else
-              python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+              python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
             fi
 
             pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple

.github/workflows/_logprob_test_linux.yml

@@ -134,7 +134,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_pre_ce_test.yml

@@ -154,7 +154,7 @@ jobs:
           --gpus "\"device=${DEVICES}\"" ${docker_image} /bin/bash -c '
           git config --global --add safe.directory /workspace/FastDeploy
           cd FastDeploy
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
           python -m pip install ${fd_wheel_url}
           bash scripts/run_pre_ce.sh
           '

.github/workflows/_stable_test.yml

@@ -146,7 +146,7 @@ jobs:
           -v "${CACHE_DIR}/ConfigDir:/root/.config" \
           -e TZ="Asia/Shanghai" \
           --gpus '"device='"${DEVICES}"'"' ${docker_image} /bin/bash -xc '
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 

.github/workflows/_unit_test_coverage.yml

@@ -175,7 +175,7 @@ jobs:
           git config --global --add safe.directory /workspace/FastDeploy
           cd FastDeploy
           git diff origin/${BASE_REF}..HEAD --unified=0 > diff.txt
-          python -m pip install paddlepaddle-gpu==3.3.0.dev20251118 -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
+          python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
           pip config set global.extra-index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 
           python -m pip install -r scripts/unittest_requirement.txt

build.sh

@@ -77,47 +77,49 @@ function copy_ops(){
     PY_VERSION="py${PY_MAIN_VERSION}.${PY_SUB_VERSION}"
     SYSTEM_VERSION=`${python} -c "import platform; print(platform.system().lower())"`
     PROCESSOR_VERSION=`${python} -c "import platform; print(platform.processor())"`
-    WHEEL_NAME="fastdeploy_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
-    WHEEL_CPU_NAME="fastdeploy_cpu_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
+    EGG_NAME="fastdeploy_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
+    EGG_CPU_NAME="fastdeploy_cpu_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
 
     # Add compatibility for modern python packaging methods
-    WHEEL_MODERN_NAME="fastdeploy_ops"
-    WHEEL_MODERN_CPU_NAME="fastdeploy_cpu_ops"
+    LEGACY_PACKAGE_DIR="${OPS_TMP_DIR}/${EGG_NAME}"
+    MODERN_PACKAGE_DIR="${OPS_TMP_DIR}/fastdeploy_ops"
+    LEGACY_PACKAGE_DIR_CPU="${OPS_TMP_DIR}/${EGG_CPU_NAME}"
+    MODERN_PACKAGE_DIR_CPU="${OPS_TMP_DIR}/fastdeploy_cpu_ops"
 
-    # Handle GPU ops directories (WHEEL_MODERN_NAME and WHEEL_NAME)
-    if [ -d "./${OPS_TMP_DIR}/${WHEEL_MODERN_NAME}" ]; then
+    # Handle GPU ops directory compatibility between modern and legacy naming
+    if [ -d "${MODERN_PACKAGE_DIR}" ]; then
         echo -e "${GREEN}[Info]${NONE} Ready to copy ops from modern directory ${WHEEL_MODERN_NAME} to target directory"
-        # Set WHEEL_NAME to empty string to ignore the directory path
-        WHEEL_NAME=""
+        TMP_PACKAGE_DIR="${OPS_TMP_DIR}"
+    # If modern directory doesn't exist, check for legacy directory, this branch should be removed in the future
+    elif [ -d "${LEGACY_PACKAGE_DIR}" ]; then
+        echo -e "${YELLOW}[Warning]${NONE} ${EGG_NAME} directory exists. This is a legacy packaging and distribution method."
+        TMP_PACKAGE_DIR="${LEGACY_PACKAGE_DIR}"
     else
-        # If modern directory doesn't exist, check for deprecated directory
-        if [ -d "./${OPS_TMP_DIR}/${WHEEL_NAME}" ]; then
-            echo -e "${YELLOW}[Warning]${NONE} ${WHEEL_NAME} directory exists. This is a deprecated packaging and distribution method."
-        fi
+        echo -e "${RED}[Error]${NONE} Neither modern nor legacy directory for gpu ops found in ${OPS_TMP_DIR}"
     fi
 
-    # Handle CPU ops directories (WHEEL_MODERN_CPU_NAME and WHEEL_CPU_NAME)
-    if [ -d "./${OPS_TMP_DIR}/${WHEEL_MODERN_CPU_NAME}" ]; then
+    # Handle CPU ops directory compatibility between modern and legacy naming
+    if [ -d "${MODERN_PACKAGE_DIR_CPU}" ]; then
         echo -e "${GREEN}[Info]${NONE} Ready to copy ops from modern directory ${WHEEL_MODERN_CPU_NAME} to target directory"
-        # Set WHEEL_CPU_NAME to empty string to ignore the directory path
-        WHEEL_CPU_NAME=""
+        TMP_PACKAGE_DIR_BASE="${OPS_TMP_DIR}"
+    # If modern directory doesn't exist, check for legacy directory, this branch should be removed in the future
+    elif [ -d "${LEGACY_PACKAGE_DIR_CPU}" ]; then
+        echo -e "${YELLOW}[Warning]${NONE} ${EGG_CPU_NAME} directory exists. This is a legacy packaging and distribution method."
+        TMP_PACKAGE_DIR_BASE="${LEGACY_PACKAGE_DIR_CPU}"
     else
-        # If modern directory doesn't exist, check for deprecated directory
-        if [ -d "./${OPS_TMP_DIR}/${WHEEL_CPU_NAME}" ]; then
-            echo -e "${YELLOW}[Warning]${NONE} ${WHEEL_CPU_NAME} directory exists. This is a deprecated packaging and distribution method."
-        fi
+        echo -e "${YELLOW}[Warning]${NONE} Neither modern nor legacy directory for cpu ops found in ${OPS_TMP_DIR}"
     fi
     is_rocm=`$python -c "import paddle; print(paddle.is_compiled_with_rocm())"`
     if [ "$is_rocm" = "True" ]; then
       DEVICE_TYPE="rocm"
-      cp -r ./${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/gpu
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/gpu
       echo -e "ROCM ops have been copy to fastdeploy"
       return
     fi
     is_cuda=`$python -c "import paddle; print(paddle.is_compiled_with_cuda())"`
     if [ "$is_cuda" = "True" ]; then
       DEVICE_TYPE="gpu"
-      cp -r ./${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/gpu
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/gpu
       echo -e "CUDA ops have been copy to fastdeploy"
       return
     fi
@@ -125,7 +127,7 @@ function copy_ops(){
     is_xpu=`$python -c "import paddle; print(paddle.is_compiled_with_xpu())"`
     if [ "$is_xpu" = "True" ]; then
       DEVICE_TYPE="xpu"
-      cp -r ./${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/xpu
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/xpu
       echo -e "xpu ops have been copy to fastdeploy"
       return
     fi
@@ -133,7 +135,7 @@ function copy_ops(){
     is_npu=`$python -c "import paddle; print(paddle.is_compiled_with_custom_device('npu'))"`
     if [ "$is_npu" = "True" ]; then
       DEVICE_TYPE="npu"
-      cp -r ${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/npu
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/npu
       echo -e "npu ops have been copy to fastdeploy"
       return
     fi
@@ -141,7 +143,7 @@ function copy_ops(){
     if_corex=`$python -c "import paddle; print(paddle.is_compiled_with_custom_device(\"iluvatar_gpu\"))"`
     if [ "$if_corex" = "True" ]; then
       DEVICE_TYPE="iluvatar-gpu"
-      cp -r ./${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/iluvatar
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/iluvatar
       echo -e "Iluvatar ops have been copy to fastdeploy"
       return
     fi
@@ -149,7 +151,7 @@ function copy_ops(){
     is_gcu=`$python -c "import paddle; print(paddle.is_compiled_with_custom_device('gcu'))"`
     if [ "$is_gcu" = "True" ]; then
       DEVICE_TYPE="gcu"
-      cp -r ${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/gcu
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/gcu
       echo -e "gcu ops have been copy to fastdeploy"
       return
     fi
@@ -158,8 +160,8 @@ function copy_ops(){
     if [ "$is_maca" = "True" ]; then
       DEVICE_TYPE="metax_gpu"
       mkdir -p ../fastdeploy/model_executor/ops/base
-      cp -r ./${OPS_TMP_DIR_BASE}/${WHEEL_BASE_NAME}/* ../fastdeploy/model_executor/ops/base
-      cp -r ./${OPS_TMP_DIR}/${WHEEL_NAME}/* ../fastdeploy/model_executor/ops/gpu
+      cp -r ${OPS_TMP_DIR_BASE}/${WHEEL_BASE_NAME}/* ../fastdeploy/model_executor/ops/base
+      cp -r ${TMP_PACKAGE_DIR}/* ../fastdeploy/model_executor/ops/gpu
       echo -e "MACA ops have been copy to fastdeploy"
       return
     fi
@@ -249,8 +251,14 @@ function extract_ops_from_precompiled_wheel() {
   echo -e "${BLUE}[precompiled]${NONE} Copying GPU precompiled contents..."
   mkdir -p "$DST_DIR"
   cp -r "$SRC_DIR/deep_gemm" "$DST_DIR/" 2>/dev/null || true
-  cp -r "$SRC_DIR/fastdeploy_ops.py" "$DST_DIR/" 2>/dev/null || true
-  cp -f "$SRC_DIR/"fastdeploy_ops_*.so "$DST_DIR/" 2>/dev/null || true
+  # Check for modern Python packaging approach (fastdeploy_ops directory)
+  # If exists, copy the entire directory; otherwise, fall back to legacy method (individual files)
+  if [ -d "$SRC_DIR/fastdeploy_ops" ]; then
+    cp -r "$SRC_DIR/fastdeploy_ops" "$DST_DIR/" 2>/dev/null || true
+  else
+    cp -r "$SRC_DIR/fastdeploy_ops.py" "$DST_DIR/" 2>/dev/null || true
+    cp -f "$SRC_DIR/"fastdeploy_ops_*.so "$DST_DIR/" 2>/dev/null || true
+  fi
   cp -f "$SRC_DIR/version.txt" "$DST_DIR/" 2>/dev/null || true
 
   echo -e "${GREEN}[SUCCESS]${NONE} Installed FastDeploy using precompiled wheel."

custom_ops/xpu_ops/build.sh

@@ -24,30 +24,39 @@ PY_SUB_VERSION=`${python} -V 2>&1 | awk '{print $2}' | awk -F '.' '{print $2}'`
 PY_VERSION="py${PY_MAIN_VERSION}.${PY_SUB_VERSION}"
 SYSTEM_VERSION=`${python} -c "import platform; print(platform.system().lower())"`
 PROCESSOR_VERSION=`${python} -c "import platform; print(platform.processor())"`
-WHEEL_NAME="fastdeploy_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
+EGG_NAME="fastdeploy_ops-${OPS_VERSION}-${PY_VERSION}-${SYSTEM_VERSION}-${PROCESSOR_VERSION}.egg"
 
 # Add compatibility for modern python packaging methods
-WHEEL_MODERN_NAME="fastdeploy_ops"
+LEGACY_PACKAGE_DIR="${OPS_TMP_DIR}/${EGG_NAME}"
+MODERN_PACKAGE_DIR="${OPS_TMP_DIR}/fastdeploy_ops"
+
+# Check if OPS_TMP_DIR exists, create it if it doesn't
+if [ ! -d "${OPS_TMP_DIR}" ]; then
+    echo -e "${GREEN}[Info]${NONE} Creating directory ${OPS_TMP_DIR}"
+    mkdir -p "${OPS_TMP_DIR}"
+fi
 
 ${python} setup_ops.py install --install-lib ${OPS_TMP_DIR}
 
 # Handle directory compatibility between modern and legacy naming
-if [ -d "./${OPS_TMP_DIR}/${WHEEL_MODERN_NAME}" ]; then
-    echo -e "${GREEN}[Info]${NONE} Ready to use ops from modern directory ${WHEEL_MODERN_NAME}"
+if [ -d "${MODERN_PACKAGE_DIR}" ]; then
+    echo -e "${GREEN}[Info]${NONE} Ready to use ops from modern directory ${MODERN_PACKAGE_DIR}"
     # Use modern directory name
-    TARGET_DIR="${OPS_TMP_DIR}/${WHEEL_MODERN_NAME}"
-else
-    # If modern directory doesn't exist, check for legacy directory
-    if [ -d "./${OPS_TMP_DIR}/${WHEEL_NAME}" ]; then
-        echo -e "${YELLOW}[Warning]${NONE} ${WHEEL_NAME} directory exists. This is a deprecated packaging and distribution method."
-    else
-        echo -e "${RED}[Error]${NONE} Neither modern nor legacy directory found in ${OPS_TMP_DIR}"
-    fi
+    TMP_PACKAGE_DIR="${OPS_TMP_DIR}"
+    CUSTOM_OP_DLL_RPATH='$ORIGIN/../libs'
+    CUSTOM_OP_DLL_PATH="${MODERN_PACKAGE_DIR}/fastdeploy_ops_pd_.so"
+# If modern directory doesn't exist, check for legacy directory, this branch should be removed in the future
+elif [ -d "${LEGACY_PACKAGE_DIR}" ]; then
+    echo -e "${YELLOW}[Warning]${NONE} ${LEGACY_PACKAGE_DIR} directory exists. This is a deprecated packaging and distribution method."
     # Use legacy directory name
-    TARGET_DIR="${OPS_TMP_DIR}/${WHEEL_NAME}"
+    TMP_PACKAGE_DIR="${LEGACY_PACKAGE_DIR}"
+    CUSTOM_OP_DLL_RPATH='$ORIGIN/libs'
+    CUSTOM_OP_DLL_PATH="${TMP_PACKAGE_DIR}/fastdeploy_ops_pd_.so"
+else
+    echo -e "${RED}[Error]${NONE} Neither modern nor legacy directory for xpu ops found in ${OPS_TMP_DIR}"
 fi
 
-mkdir -p ${TARGET_DIR}/libs
-cp ${XVLLM_PATH}/xft_blocks/so/libxft_blocks.so ${TARGET_DIR}/libs/
-cp ${XVLLM_PATH}/infer_ops/so/libapiinfer.so ${TARGET_DIR}/libs/
-patchelf --set-rpath '$ORIGIN/libs' ${TARGET_DIR}/fastdeploy_ops_pd_.so
+mkdir -p ${TMP_PACKAGE_DIR}/libs
+cp ${XVLLM_PATH}/xft_blocks/so/libxft_blocks.so ${TMP_PACKAGE_DIR}/libs/
+cp ${XVLLM_PATH}/infer_ops/so/libapiinfer.so ${TMP_PACKAGE_DIR}/libs/
+patchelf --set-rpath ${CUSTOM_OP_DLL_RPATH} ${CUSTOM_OP_DLL_PATH}

scripts/run_ci_xpu.sh

@@ -42,7 +42,7 @@ echo "uninstall org"
 python -m pip uninstall paddlepaddle-xpu -y
 python -m pip uninstall fastdeploy-xpu -y
 
-python -m pip install https://paddle-whl.bj.bcebos.com/nightly/xpu-p800/paddlepaddle-xpu/paddlepaddle_xpu-3.3.0.dev20251118-cp310-cp310-linux_x86_64.whl
+python -m pip install paddlepaddle-xpu -i https://www.paddlepaddle.org.cn/packages/nightly/xpu-p800/
 
 echo "build whl"
 bash custom_ops/xpu_ops/download_dependencies.sh develop

setup.py

@@ -275,20 +275,31 @@ setup(
     url="https://github.com/PaddlePaddle/FastDeploy",
     packages=find_packages(),
     package_dir={"fastdeploy": "fastdeploy/"},
+    # For deprecated method (egg-based installation), `.so` files are placed in the `model_executor/ops/XXX` directory.
+    # For modern method (PEP 517/518-based installation), `.so` files are placed in the `model_executor/ops/XXX/fastdeploy_ops` directory.
+    # Therefore, the `fastdeploy_ops` directory should be included for modern Python packaging.
     package_data={
         "fastdeploy": [
             "model_executor/ops/gpu/*",
+            "model_executor/ops/gpu/fastdeploy_ops/*",
             "model_executor/ops/gpu/deep_gemm/include/**/*",
             "model_executor/ops/cpu/*",
+            "model_executor/ops/cpu/fastdeploy_cpu_ops/*",
             "model_executor/ops/xpu/*",
+            "model_executor/ops/xpu/fastdeploy_ops/*",
             "model_executor/ops/xpu/libs/*",
+            "model_executor/ops/xpu/fastdeploy_ops/libs/*",
             "model_executor/ops/npu/*",
+            "model_executor/ops/npu/fastdeploy_ops/*",
             "model_executor/ops/base/*",
+            "model_executor/ops/base/fastdeploy_ops/*",
             "model_executor/ops/iluvatar/*",
+            "model_executor/ops/iluvatar/fastdeploy_ops/*",
             "model_executor/models/*",
             "model_executor/layers/*",
             "input/ernie4_5_vl_processor/utils/*",
             "model_executor/ops/gcu/*",
+            "model_executor/ops/gcu/fastdeploy_ops/*",
             "version.txt",
         ]
     },