MLModelScope: A Distributed Platform for Model Evaluation and Benchmarking at Scale

TL;DR

MLModelScope is an open-source, scalable platform that simplifies and standardizes the evaluation and benchmarking of machine learning models across diverse frameworks and hardware, enabling fair and repeatable comparisons.

Contribution

It introduces a distributed, framework-agnostic platform for scalable, customizable, and fair ML/DL model evaluation and benchmarking with multi-interface support.

Findings

Supports all major ML frameworks and hardware

Enables parallel evaluation for efficiency

Demonstrates impact of pipeline and hardware choices

Abstract

Machine Learning (ML) and Deep Learning (DL) innovations are being introduced at such a rapid pace that researchers are hard-pressed to analyze and study them. The complicated procedures for evaluating innovations, along with the lack of standard and efficient ways of specifying and provisioning ML/DL evaluation, is a major "pain point" for the community. This paper proposes MLModelScope, an open-source, framework/hardware agnostic, extensible and customizable design that enables repeatable, fair, and scalable model evaluation and benchmarking. We implement the distributed design with support for all major frameworks and hardware, and equip it with web, command-line, and library interfaces. To demonstrate MLModelScope's capabilities we perform parallel evaluation and show how subtle changes to model evaluation pipeline affects the accuracy and HW/SW stack choices affect performance.