AIMET Installation

Release packages

AIMET release packages are hosted at https://github.com/quic/aimet/releases. Each release includes multiple python packages of the following format:

<PACKAGE_PREFIX>-<VARIANT>_<VERSION>-cp38-cp38-linux_x86_64.whl

Please find more information below about each VARIANT.

PyTorch

  1. torch-gpu for PyTorch 1.13 GPU package with Python 3.8 and CUDA 11.x - Recommended for use with PyTorch models

  2. torch-cpu for PyTorch 1.13 CPU package with Python 3.8 - If installing on a machine without CUDA

  3. torch-gpu-pt19 for PyTorch 1.9 GPU package with Python 3.8 and CUDA 11.x

  4. torch-cpu-pt19 for PyTorch 1.9 CPU package with Python 3.8 - If installing on a machine without CUDA

TensorFlow

  1. tf-gpu for TensorFlow 2.10 GPU package with Python 3.8 - Recommended for use with TensorFlow models

  2. tf-cpu for TensorFlow 2.10 CPU package with Python 3.8 - If installing on a machine without CUDA

ONNX

  1. onnx-gpu for ONNX 1.11.0 GPU package with Python 3.8 - Recommended for use with ONNX models

  2. onnx-cpu for ONNX 1.11.0 CPU package with Python 3.8 - If installing on a machine without CUDA

System Requirements

The AIMET package requires the following host platform setup:

  • 64-bit Intel x86-compatible processor

  • Linux Ubuntu: 20.04 LTS

  • bash command shell

  • For GPU variants:

To use the GPU accelerated training modules an Nvidia CUDA enabled GPU with a minimum Nvidia driver version of 455+ is required. Using the latest driver is always recommended, especially if using a newer GPU. Both CUDA and cuDNN (the more advanced CUDA interface) enabled GPUs are supported.

Recommended host system hardware requirements:

  • Intel i7 multicore CPU w/hyperthreading

  • 16+ GB RAM

  • 500GB+ SSD hard drive

  • For GPU variants:

    • GPU: Nvidia GeForce GTX 1080 or Tesla V100

While these are not minimum requirements, they are recommended for good performance when training large networks.

Installation Instructions

There are two ways to setup and install AIMET:

  • On your host machine

  • Using our pre-built development Docker images

Please click on the appropriate link for installation instructions: