Skip to content
BENIBIX Learn

MLOps & Deployment

Training, serving and running models in production · 25 terms

Sort
Compare

MLOps Basics

9

What it means to run models like software. Also the tools that make training easy to do again.

The tools and habits that keep an AI model working. They start once people use it.

How teams run apps built on language models. It covers prompts, tests and cost.

A chain of steps that turns raw data into a model. Code runs it.

Free tools that run machine learning steps on Kubernetes. Each step gets its own box.

Notes on each training run. You can then compare them and redo the best one.

A free tool that logs each training run and stores the models it makes.

A web tool that charts each training run live. A team can then compare them.

One shared store of model inputs. Training and the live app use the same ones.

Software that tries many models on your data. It keeps the one that scores best.

Large-Scale Training

4

Training one big model on many GPUs at once, and saving its progress so a crash does not lose it.

Training one model on many GPUs at once. A big job then takes days, not years.

Each GPU holds the whole model. It trains on its own slice of the data.

A model too big for one GPU is cut into parts. Each GPU holds one part.

A saved copy of a model partway through training. A crash then loses little.

Model Deployment

7

Getting a trained model out to real users. Storing it, serving it and running it.

A list of each model version, its score, and which one is live.

A model put behind a web address. Apps send it data and get answers.

A model runs on lots of data at set times. The results are saved.

A model runs the moment a request comes in. You wait for the answer.

An open file type for AI models. A model made in one tool runs in others.

Ways to make a trained model answer faster and cheaper. Its answers stay almost as good.

A new model gets copies of real requests. Its answers are saved but not shown.

Production Monitoring

5

Watching a live model, spotting when the world has changed, and training it again.

Keeping an eye on a live model. You catch problems before they do harm.

The data that comes in starts to look less like what the model learned on.

The right answer for the same input changes. So a model's rules go out of date.

Live data and training data are made in two ways. So a good model does badly.

Training a model again on new data. It keeps the model up to date.