Framework finds best way to run AI models on flexible accelerators
qGAMMA: A Framework for Optimal DNN Mapping on Multi-Precision Accelerators
Hardware ArchitectureArtificial Intelligence
Summary
Running AI models efficiently can be tricky because they use different types of numbers. The authors created qGAMMA, a tool that helps figure out the best way to run these models on special computer chips that handle many number types. It uses smart searching and knowledge about the hardware to find the fastest and most efficient setup. This helps balance speed, energy use, and hardware limits when running AI.
What this means in practice
- •For ai hardware designers: Find optimal configurations for hardware running mixed-precision neural networks to improve speed and energy efficiency.
- •For embedded systems developers: Adjust neural network deployment on resource-constrained devices to better trade off accuracy, speed, and power consumption.
Authors
Leandro Fiorin, Luigi Altamura, Cristina Silvano
Abstract
This work describes qGAMMA, a flexible framework for exploring and optimizing the mapping of mixed-precision Deep Neural Network (DNN) models onto general-purpose spatial accelerators that support a range of fixed- and floating-point precisions. qGAMMA navigates the design space using a domain-specific genetic algorithm and an analytical performance model based on hardware synthesis results. The framework identifies the optimal mapping and accelerator configuration that minimizes target metrics. Extensive evaluations across DNN workloads show that qGAMMA offers valuable insights for designing mixed-precision applications, enabling better trade-offs between performance, energy efficiency, and resource utilization.