Install Intel® AI Analytics Toolkit via Conda*
Intel provides access to the AI Kit through a public Anaconda repository. If you do not have an existing Conda-based python environment, install Conda or Miniconda*. To get more details on the AI Analytics Toolkit, visit the Intel AI Analytics toolkit home page. To test AI Kit conda metapackages functionality, conda-libmamba-solver is used.
NOTE:
The July 2023 editions of the Anaconda Distribution and miniconda include the conda-libmamba-solver package in the base environment. To switch to the libmamba solver, add --solver=libmamba (for one-time runs) or conda-config --set solver=libmamba (for persistent switching via the .condarc file). For more details, refer to the conda-libmamba-solver Getting Started Guide. Starting September 2023, the default conda solver is libmamba.
The AI Kit contains the following distinct Python environments targeting different use cases:
intel-aikit-tensorflow for deep learning workflows using Intel® Optimization for TensorFlow*.
intel-aikit-tensorflow-gpu for deep learning workflows using Intel® Extension for TensorFlow* that provides Intel GPU support
intel-aikit-pytorch for deep learning workflows using Intel® Optimization for PyTorch* and Intel® Extension for PyTorch*
intel-aikit-pytorch-gpu for deep learning workflows using Intel® Optimization for PyTorch* and Intel® Extension for PyTorch* that provides Intel GPU support
intel-aikit-modin for data analytics and machine learning workflows using Intel® Distribution of Modin (for accelerated Panda data frames), Intel® Extension for Scikit-learn*, and Intel optimizations for XGboost (for ML training and inference).
intel-aikit for data science workstation development. The oneAPI AI kit samples and documents are not applicable to the intel-aikit Conda package.
NOTE:
To get the latest version of the Intel(R) Optimization for TensorFlow*, you must first install Python 3.9, then install the AI Kit through Anaconda.
To instal the AI Kit via Conda, complete the following steps:
Activate your existing conda environment located in <pythonhome>:
source <conda-install-dir>/bin/activate
Install the AI Kit oneAPI packages in a new environment using conda create. A list of available packages is located at https://anaconda.org/intel/repo. Not all packages in the Anaconda repository are up to date with the current release. If the repo contains an outdated version of a required component, get a newer one by installing via the command line or GUI.
If the repository contains the desired version, create the corresponding environment using the commands below:
AI Kit TensorFlow* aikit-tf
conda create -n aikit-tf -c intel -c conda-forge --override-channels python=3.9 intel-aikit-tensorflow=2023.2.0
AI Kit TensorFlow* with Intel GPU support aikit-tf-gpu
conda create -n aikit-tf-gpu -c intel -c conda-forge --override-channels python=3.9 intel-aikit-tensorflow-gpu=2023.2.0
AI Kit PyTorch* aikit-pt
conda create -n aikit-pt -c intel -c conda-forge --override-channels python=3.9 intel-aikit-pytorch=2023.2.0
AI Kit PyTorch* with Intel GPU support aikit-pt-gpu
conda create -n aikit-pt-gpu -c intel -c conda-forge --override-channels python=3.9 intel-aikit-pytorch-gpu=2023.2.0
AI Kit Modin and machine learning aikit-modin
conda create -n aikit-modin -c intel -c conda-forge intel-aikit-modin
Set user environment. After the toolkit is installed, before accessing the tools, you must activate your python environment and set up environment variables to access the tools. For example, to activate the python environment created in the previous step, use:
conda activate aikit-tf
NOTE:
- To install the Model Zoo for Intel® Architecture component of the toolkit, clone the main branch to your local directory: git clone https://github.com/IntelAI/models.git.
If you have applications with long-running GPU compute workloads in native environments, you must disable the hangcheck timeout period to avoid terminating workloads.
List of Available Packages
NOTE:
Intel® packages are available on intel label on the Anaconda* Cloud. The necessary dependencies for Intel® packages can be found on the conda-forge channel. You must include -c intel -c conda-forge on your command line as in the examples above, or add both channels to your Conda configuration file using:
conda config --add channels conda-forge conda config --add channels intel
Component Name |
Package Name |
Platform |
|---|---|---|
Intel® Distribution for Python* |
intelpython3_full |
linux-x64 |
Intel® Distribution of Modin* (via Anaconda distribution of the toolkit using the Conda package manager) |
intel-aikit-modin |
linux-x64 |
Intel® Neural Compressor |
neural-compressor |
linux-x64 |
Intel® Optimization for PyTorch* |
intel-aikit-pytorch (for CPU only systems), intel-aikit-pytorch-gpu (for systems with GPU) |
linux-x64 |
Intel® Optimization for TensorFlow* |
intel-aikit-tensorflow (for CPU only systems), intel-aikit-tensorflow-gpu (for systems with GPU) |
linux-x64 |
After you have installed your components, view the Get Started Guide for the Intel oneAPI AI Analytics Toolkit to build and run a sample or explore Getting Started Samples on GitHub.