Date of Award

2025-12-01

Degree Name

Master of Science

Department

Computer Science

Advisor(s)

Shirley V. Moore

Abstract

This thesis presents a tool to profile deep learning (DL) and machine learning (ML) models by collecting FLOPs, memory movement, and timing data through cyPAPI to generate roofline performance models. The tool is containerized for portability and reproducibility, integrates directly with PyTorch workflows, and provides fine grained insights into computational bottlenecks across model components. Unlike prior system-level or benchmarking-centric tools, this project empowers developers and researchers with an accessible, modular framework for performance analysis and optimization.

Language

en

Provenance

Received from ProQuest

File Size

61 p.

File Format

application/pdf

Rights Holder

Irvin Lopez-Audetat

Share

COinS