Timing Predictability with IREE

This component is a demonstrator showing how to use the IREE framework to compile a model on a specific target to master the inference process.
At present, the IREE framework cannot work with any architecture, nor can it handle all types of model.
It also requires the models to be tested to be on a server accessible directly by url.
Various models are already available on the platform
It is theoretically possible to test tensorflow, tensorflow lite and pytorch models.
Onnx and pytorch formats are also possible, although a format conversion must be performed beforehand to be readable by IREE.
The procedure below shows how to install and configure the platform to load and test an inference model on a specific architecture, namely a Google pixel 6 smartphone.

Documentation

Methodological Guidelines

• Performance Analysis Guidelines
• Performance prediction definition

Benchmark

• IREE performance analysis
• IREE Study on MLIR Based end to end compiler
• Timing Predictability and Repeatability

State of the Art

• Definition of the Study Area
• Embedded AI components State of the art
• Methods and Tools for the Optimization of Machine Learning Components SOTA

Support

Support for Timing Predictability with IREE must be obtained by sending an email to support@confiance.ai

Ensure your email contains :

• Your name
• A link to this page
• the version you are working with
• A clear description of the problematic (bug, crash, feature or help request)
• A full description of the problem whichallow to reproduce it
• Any file or screenshort element mandatory for the full understanding of the problematic