All Products
Search

This Product

    Platform For AI:Install PAI-Blade

    Document Center

    Platform For AI:Install PAI-Blade

    Last Updated:Mar 06, 2026

    PAI-Blade accelerates deep learning inference on Platform for AI (PAI). The installation package includes a Python wheel package for model optimization and a C++ SDK for model inference.

    Prerequisites

    Ensure the following requirements are met:

    • Linux operating system

    • Python 3.6, 3.7, or 3.8

    • Supported deep learning framework installed (TensorFlow or PyTorch)

    • GPU environments only: CUDA 10.0 to 11.3

    Compatibility

    Category Supported versions
    Operating system Linux
    Python 3.6, 3.7, 3.8
    CUDA 10.0 to 11.3 (GPU environments)
    TensorFlow 1.15, 2.4, 2.7
    PyTorch 1.6.0 and later
    TensorRT 8.0 and later
    C++ SDK ABI cxx11, pre-cxx11
    C++ SDK formats RPM, DEB, TGZ
    Device types GPU, CPU, terminal devices (Mobile Neural Network (MNN))

    Installation paths

    Scenario Components When to use
    GPU or CPU server Framework + TensorRT + wheel package + C++ SDK + access token Model optimization and inference on servers
    On-premises device TensorFlow + MNN + wheel package Model optimization for mobile or embedded devices

    Install on a GPU or CPU server

    Step 1: Install the framework

    Install TensorFlow or PyTorch before installing PAI-Blade. PAI-Blade does not install frameworks automatically.

    TensorFlow

    PAI-Blade supports TensorFlow 1.15, 2.4, and 2.7. Install from the TensorFlow community:

    # GPU
pip3 install tensorflow-gpu==1.15.0
pip3 install tensorflow-gpu==2.4.0

# CPU
pip3 install tensorflow==1.15.0
pip3 install tensorflow==2.4.0

    For TensorFlow with integrated TensorRT, use the precompiled package from PAI instead of the community package.

    PyTorch

    PAI-Blade supports PyTorch 1.6.0 and later. Install from the PyTorch official website based on your device type and CUDA version.

    Example: Install PyTorch 1.7.1 with CUDA 11.0:

    pip3 install torch==1.7.1+cu110 torchvision==0.8.2+cu110 \
    -f https://download.pytorch.org/whl/torch/

    Official PyTorch 1.6.0 does not support CUDA 10.0. Use the precompiled PyTorch package from PAI for this combination.

    Step 2: Install the wheel package

    Select the pip3 install command matching your framework, framework version, device type, and CUDA version.

    For earlier versions, see Installation commands and SDK download URLs for PAI-Blade of earlier versions.

    CPU

    TensorFlow 1.15.0 and PyTorch 1.6.0

    pip3 install pai_blade_cpu==3.27.0+1.15.0.1.6.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install tensorflow_blade_cpu==3.27.0+1.15.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install torch_blade_cpu==3.27.0+1.6.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    TensorFlow 2.4.0 and PyTorch 1.7.1

    pip3 install pai_blade_cpu==3.27.0+2.4.0.1.7.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install tensorflow_blade_cpu==3.27.0+2.4.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install torch_blade_cpu==3.27.0+1.7.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    PyTorch 1.8.1

    pip3 install pai_blade_cpu==3.27.0+1.8.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install torch_blade_cpu==3.27.0+1.8.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    PyTorch 1.9.0

    pip3 install pai_blade_cpu==3.27.0+1.9.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install torch_blade_cpu==3.27.0+1.9.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    TensorFlow 2.7.0 and PyTorch 1.10.0

    pip3 install pai_blade_cpu==3.27.0+2.7.0.1.10.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install tensorflow_blade_cpu==3.27.0+2.7.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
pip3 install torch_blade_cpu==3.27.0+1.10.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    CUDA 11.0

    TensorFlow 2.4.0 and PyTorch 1.7.1

    pip3 install pai_blade_gpu==3.27.0+cu110.2.4.0.1.7.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install tensorflow_blade_gpu==3.27.0+cu110.2.4.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.7.1.cu110 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    CUDA 11.1

    PyTorch 1.8.1

    pip3 install pai_blade_gpu==3.27.0+cu111.1.8.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.8.1.cu111 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    PyTorch 1.9.0

    pip3 install pai_blade_gpu==3.27.0+cu111.1.9.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.9.0.cu111 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    PyTorch 1.10.0

    pip3 install pai_blade_gpu==3.27.0+cu111.1.10.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.10.0.cu111 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    CUDA 11.2

    TensorFlow 2.7.0

    pip3 install pai_blade_gpu==3.27.0+cu112.2.7.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install tensorflow_blade_gpu==3.27.0+cu112.2.7.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    CUDA 11.3

    PyTorch 1.11.0

    pip3 install pai_blade_gpu==3.27.0+cu113.1.11.0 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.11.0.cu113 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    PyTorch 1.12.1

    pip3 install pai_blade_gpu==3.27.0+cu113.1.12.1 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html
pip3 install torch_blade==3.27.0+1.12.1.cu113 -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo_ext.html

    Step 3: Install the C++ SDK

    The C++ SDK supports only the GNU Compiler Collection (GCC) on Linux. Blade provides packages with two Application Binary Interface (ABI) variants. For details, see the GCC ABI documentation.

    Select your ABI variant

    ABI variant When to use
    pre-cxx11 GCC earlier than 5.1, or _GLIBCXX_USE_CXX11_ABI=0 is configured
    cxx11 GCC 5.1 or later without _GLIBCXX_USE_CXX11_ABI=0

    Select your package format

    Format Distribution Install command Root required
    RPM CentOS, Red Hat rpm -ivh Yes
    DEB Ubuntu, Debian dpkg -i Yes
    TGZ Any Linux Extract and use No

    Example: Install the pre-cxx11 ABI SDK for CUDA 11.0 (v3.23.0)

    RPM

    rpm -ivh https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/3.23.0/py3.6.8_cu110_tf2.4.0_torch1.7.1_abiprecxx11/blade_cpp_sdk_gpu-3.23.0-Linux.rpm

    DEB

    wget https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/3.23.0/py3.6.8_cu110_tf2.4.0_torch1.7.1_abiprecxx11/blade_cpp_sdk_gpu-3.23.0-Linux.deb
dpkg -i blade_cpp_sdk_gpu-3.23.0-Linux.deb

    SDK directory structure

    RPM and DEB packages install to /usr/local/ by default. Directory structure after installation:

    /usr/local/
├── bin
│   ├── disc_compiler_main
│   └── tao_compiler_main
└── lib
    ├── libral_base_context.so
    ├── libtao_ops.so
    ├── libtf_blade.so
    ├── libtorch_blade.so
    └── mlir_disc_builder.so

    Use the dynamic-link libraries in /usr/local/lib for model deployment.

    SDK download URLs (v3.27.0)

    For earlier versions, see Installation commands and SDK download URLs for PAI-Blade of earlier versions.

    cxx11 ABI

    Device Framework versions DEB RPM TGZ
    CPU TensorFlow 1.15.0 + PyTorch 1.6.0 DEB RPM TGZ

    pre-cxx11 ABI

    Device Framework versions DEB RPM TGZ
    CPU TensorFlow 1.15.0 + PyTorch 1.6.0 DEB RPM TGZ
    CPU TensorFlow 2.4.0 + PyTorch 1.7.1 DEB RPM TGZ
    CPU PyTorch 1.8.1 DEB RPM TGZ
    CPU PyTorch 1.9.0 DEB RPM TGZ
    CPU TensorFlow 2.7.0 + PyTorch 1.10.0 DEB RPM TGZ
    CUDA 11.0 TensorFlow 2.4.0 + PyTorch 1.7.1 DEB RPM TGZ
    CUDA 11.1 PyTorch 1.8.1 DEB RPM TGZ
    CUDA 11.1 PyTorch 1.9.0 DEB RPM TGZ
    CUDA 11.1 PyTorch 1.10.0 DEB RPM TGZ
    CUDA 11.2 TensorFlow 2.7.0 DEB RPM TGZ
    CUDA 11.3 PyTorch 1.11.0 DEB RPM TGZ
    CUDA 11.3 PyTorch 1.12.1 DEB RPM TGZ

    Step 4: Get an access token

    The C++ SDK requires an access token. Contact your sales manager (SA or PDSA) or join the DingTalk group for Blade users (Group ID: 21946131).

    Install for on-premises devices

    Blade converts TensorFlow models to Mobile Neural Network (MNN) format for on-device inference.

    1. Install the required frameworks:

         pip3 install tensorflow==1.15 MNN==1.1.0

    2. Install the Blade wheel package for your environment: GPU CPU

         pip3 install pai-blade-gpu \
     -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html
         pip3 install pai-blade-cpu \
     -f https://pai-blade.oss-cn-zhangjiakou.aliyuncs.com/release/repo.html

    Important notes

    • Blade does not automatically install TensorFlow or PyTorch. Install a supported framework before installing Blade.

    • Ensure the wheel package matches your device type and CUDA version.

    • Official PyTorch 1.6.0 does not support CUDA 10.0. Use the PAI-provided wheel package for this combination.