From the course: Applied AI for IT Operations (AIOps)

Unlock the full course today

Join today to access over 22,600 courses taught by industry experts or purchase this course individually.

Model serving best practices

Model serving best practices - Python Tutorial

From the course: Applied AI for IT Operations (AIOps)

Start my 1-month free trial

Model serving best practices

“

- [Instructor] One of the most critical yet overlooked area of machine learning is model serving. The ability to serve models at scale and yet in cost-effective ways is a key success factor for AI projects. Today, GPUs provide multifold processing power for machine learning. However, they are most useful in model training activities. They may or may not provide a significant advantage in model serving. It's highly recommended to compare costs between CPU and GPU implementations for serving as using 10 CPUs instead of one GPU may be cost-effective overall. Use a model serving platform like TensorFlow Serving to harness out of the box functionality and scaling. It is important to track model inference or prediction performance over time. Both accuracy and speed of prediction should be tracked to ensure that they continue to perform as desired. Model drift happens when models lose accuracy due to new types of real-time data. Model drift needs to be tracked and assessed. Pass back serving…

Contents