Overview

One of the benefits of using OpsML is its performance and simplified dependency chain compared to other python-based MLOps frameworks like MLflow. While we are continually working on adding additional benchmarks, below are some initial performance comparisons between OpsML and MLflow.

Benchmark Setup

See the benchmark setup documentation for the current benchmark setup.

For the results below, we created a separate virtual environment for each framework with only the required dependencies installed + scikit-learn. We then ran the same benchmark script for each framework to compare performance.

Benchmark Results

The following table summarizes the benchmark results comparing OpsML and MLflow for a simple model training and logging task using scikit-learn. Note - this tests both frameworks running in server mode (local storage + sqlite).

Framework	Time Taken (seconds)	Dependencies	Venv size (MB)
OpsML	0.027	9	321 (opsml - 25 (includes server deps))
MLflow	1.2	116	501