Skip to content

Latest commit

 

History

History
41 lines (23 loc) · 1.97 KB

model-zoo.md

File metadata and controls

41 lines (23 loc) · 1.97 KB

Model Zoo

Deep Java Library's (DJL) Model Zoo is more than a collection of pre-trained models. It's a bridge between a model vendor and a consumer. It provides a framework for developers to create and publish their own models.

A ZooModel has the following characteristics:

  • Globally unique: similar to Java maven packages, a model has its own group ID and artifact ID that uniquely identify it.
  • Versioned: the model version scheme allows developers to continuously update their model without causing a backward compatibility issue.
  • Ready to use out of box: the model contains predefined pre-process and post-process functionality, which allows the user to run inference with a plain java object.
  • Can be published anywhere: models can be published to an S3 bucket, a web server, or a local folder.

We provide engine agnostic ZooModels in basic model zoo package. They can be used on any DJL backend engine.

We created a Huggingface model zoo to make it easy for users to consume them.

We created a PyTorch model zoo to make it easy for users to consume them.

We created an TensorFlow model zoo to make it easy for users to consume them.

Apache MXNet has a large number of existing pre-trained models. We created an Apache MXNet model zoo to make it easy for users to consume them.

Publish your own model to the model zoo

You can create your own model in the model zoo so customers can easily consume it. For more information, see Add a new Model to the model zoo .

Load models from ModelZoo

See: How to load model