FuriosaAI SDK Installation + Prerequisites
Note
The FuriosaAI SDK is composed of a commmand line interface, runtime library, and Python library. The FuriosaAI NPU’s kernel driver, firmware, and runtime are distributed according to FuriosaAI’s evaluation tool registration and End User License Agreement. For questions about downloading and evaluating the program, please contact us at contact@furiosa.ai.
Minimum Requirements
Installing Dependencies (Linux)
To install build-essential and cmake:
$ apt-get update
$ apt-get install cmake build-essential
To install onnxruntime 1.8.1:
onnxruntime is used for ONNX model format support and model quantization.
$ wget https://github.com/hyunsik/onnxruntime/releases/download/v1.8.1/libonnxruntime-1.8.1_amd64.deb
$ apt-get install -y ./libonnxruntime-1.8.1_amd64.deb
Configuring the Python SDK Runtime Environment
Python 3.7 or later is required to use the FuriosaAI Python SDK.
Note
If you aren’t using the FuriosaAI Python SDK, you can skip this section.
python --version
Python 3.8.5
The above command checks the version of Python available on the current system. If your system doesn’t have Python or only has a lower version of it, you can configure a Python environment with one of the following methods.
Configuring the Python Environment Using Conda (recommended): Conda can be used to configure a dedicated Python environment for specific Python applications. Conda helps to prevent package dependency issues and Python version issues that can occur when installing complicated Python applications.
Configuring the Python Environment Using Linux Packages: This option is for users who have a good understanding of the Python execution environment on their Linux system and want to configure the Python environment directly on their system.
Configuring the Python Environment Using Conda
You can use Conda to configure a dedicated Python environment for specific Python applications. Please refer to the documents in Conda for more information on Conda.
Start the installation process by downloading the installation program as follows:
Run ./Miniconda3-latest-Linux-x86_64.sh and select yes.
$ wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
$ sh ./Miniconda3-latest-Linux-x86_64.sh
$ source ~/.bashrc
$ conda --version
conda 4.8.3
After installing Anaconda, you can configure an isolated Python runtime environment
and activate it as needed. The FuriosaAI Python SDK is compatible with Python 3.7-3.9 versions.
The following uses Python 3.8 version to create an execution environment named furiosa.
$ conda create -n furiosa python=3.8
The newly created Python 3.8 environment is activated with the activate command.
$ conda activate furiosa
$ python --version
Python 3.8.8
Now you can install furiosa-sdk with additional dependencies using pip as shown below. For more detailed instructions for installation, refer to FuriosaAI Python SDK Installation.
$ pip install 'furiosa-sdk[cli, runtime]'
To deactivate the use of the user created Python environment, use the deactivate command.
$ conda deactivate
Once created, the environment can be reactivated using activate and used at any time.
Packages that have already been installed do not need to be reinstalled after activation.
Configuring the Python Environment Using Linux Packages
If a Python environment can be configured directly from the system, install the following dependencies as shown below.
sudo apt install -y python3 python3-pip