This document is relevant for: Inf1, Inf2, Trn1, Trn1n

NeuronPerf 1.x Release Notes#

NeuronPerf is a lightweight Python library with a simple API that enables fast measurements of performance when running models using Neuron.

NeuronPerf release []#

Date: 3/28/2023

  • Adds trn1/inf2 support for PyTorch and TensorFlow 2.x

  • Uses new IMDSv2 for obtaining instance types

NeuronPerf release []#

Date: 11/23/2022

  • New Evaluation + metrics API (see NeuronPerf Evaluation Guide)

  • Support map and iterable-type torch datasets

  • Support custom torch DataLoader args via dataloader_kwargs

  • New get_report_by_tag utility to identify specific configurations

  • Python 3.7+ now default from 3.6

  • Pricing and sizing info updated for inf1 + trn1

Bug fixes

  • GPU inputs are now moved correctly

NeuronPerf release []#

Date: 04/29/2022

  • Minor updates

NeuronPerf release []#

Date: 03/25/2022

  • Initial release of NeuronPerf

  • Supports PyTorch, TensorFlow, and Apache MXNet (Incubating).

  • Supports customizable JSON and CSV reports

  • See NeuronPerf (Beta) for more information.

This document is relevant for: Inf1, Inf2, Trn1, Trn1n