บันทึกโมเดลลง Bento Store

BentoML is a Python framework that packages and serves models with batching and scaling built in, so you skip a lot of boilerplate. 🚀

Getting started takes one command. You install bentoml from pip just like any other Python package.

The Bento Store is a local registry on your machine. Every model you save lands there with a name and an automatic version tag.

You need a model before you can save one. Here you fit a simple scikit-learn classifier on some data.

BentoML ships a helper per framework. For scikit-learn you call bentoml.sklearn.save_model with a name and the fitted model.

Each save creates a tag like iris_clf:abc123. The name stays the same while the version after the colon changes every save.

The same pattern works everywhere. Swap the module to match your framework, like bentoml.pytorch or bentoml.xgboost.

You can see everything in the store from your terminal. The bentoml models list command shows names, tags, and sizes.

To use a saved model you load it by tag. Passing latest grabs the newest version without hardcoding a hash.

You can stash extra info at save time. Adding metadata like accuracy travels with the model so teammates know what it is.

The store can fill up over time. You prune anything you no longer need with the bentoml models delete command and a tag.

You just ran save_model twice for the same name. What identifies the two saves apart?

You installed BentoML, saved a model to the Bento Store with a framework helper, listed it, and loaded it back by tag. The model is now managed! 🙌