Version: Main branch

Community Support

The primary way in which developers will integrate and implement functionality from popular AI frameworks, is via the Predictor and Model abstractions.

The Predictor mixin class, interfaces with all AI frameworks and API providers, which provide self.predict functionality, and is subclassed by:

class	framework
`superduper.ext.sklearn.Estimator`	Scikit-Learn
`superduper.ext.transformers.Pipeline`	Hugging Face's `transformers`
`superduper.ext.torch.TorchModel`	PyTorch
`superduper.ext.openai.OpenAI`	OpenAI
`superduper.ext.cohere.Cohere`	Cohere
`superduper.ext.anthropic.Anthropic`	Anthropic
`superduper.ext.jina.Jina`	Jina

The Model class is subclassed by:

class	framework
`superduper.ext.sklearn.Estimator`	Scikit-Learn
`superduper.ext.transformers.Pipeline`	Hugging Face's `transformers`
`superduper.ext.torch.TorchModel`	PyTorch

Model instances implement self.predict, but also hold import data, such as model weights, parameters or hyperparameters. In addition, Model may implement self.fit functionality - model training and calibration.