Managing models

Manage your model's deployment and versions.
From the Models page, you can click on any model in the list to visit its overview page.

Model page

A model's name can contain any alphanumeric character, including spaces, but is not editable. The model description is shown in the Baseten UI only and is editable, just click on it.
Model overview
A model's Type like "CNN" or "NLP" refers to the category of algorithm that the model uses, while a model's Framework like "scikit-learn" or "TensorFlow" reveals the underlying technology that the model is built on. Both Type and Framework will show as "Custom" for custom models.

Model status

Each deployed version of a model has a status:
  • Deploying: While a model is being deployed, it has this intermediary status.
  • Active: The model is active and available. It can be invoked from the Python client or cURL and, if it is the primary version, from Baseten applications.
  • Deployment Failed: The model is not active. Check the logs, resolve the error, and try again. Note that deploying again will create a new version, consider using version_bump='PATCH' for fixing deployment bugs.

Model versions

Each model version counts as a model toward workspace limits.
Workspaces on the free Personal plan are limited to a single deployed model. Pre-trained models provided by Baseten do not count against this limit. To increase the number of models for your workspace, upgrade to a paid Baseten plan.
Models are managed with semantic versioning to track changes over time. Every deployment of a model with the same name as an existing model creates a new version, not a new model. Each version has a version ID, a unique string used to identify the model in the Python client and API.
Model version tab
Versions auto-increment by minor version, starting with 0.1.0. When deploying a model, use the optional argument version_bump='MAJOR' (or 'MINOR' or 'PATCH') to override this default.
While all successful deployments of a model will remain active as different versions accessible via the Python client and API, only the primary version of a model is available to applications within Baseten.
The primary version of a model will be used by default for all predictions unless you specify the desired version when calling the model. This model version is, by default, the most recent deployment of a model. You can make any active version the primary version by clicking the three-dot dropdown menu in the versions table, then selecting "Promote to primary."

Model documentation

Documentation shares essential steps and context for using a system — that's why you're reading these docs right now. Every model deployed on Baseten has an editable Readme file to document its purpose, inputs, and outputs. Models from a model template come with that information pre-filled, but you'll need to document custom models yourself.
Model use

Download the model

Any model version can be downloaded as a Truss from your Baseten account to your local computer. Just open terminal and run:
pip install --upgrade baseten
baseten login # see:
baseten models pull # asks you for a model version ID to download, find this on the model's page