Skip to main content
Version: Main branch

Model

  • Wrap a standard AI model with functionality necessary for superduper
  • Configure validation and training of a model on database data

Dependencies

(Optional dependencies)

Usage pattern

note

Note that Model is an abstract base class which cannot be called directly. To use Model you should call any of its downstream implementations, such as ObjectModel or models in the AI-integrations.

Important notes

Model instances can output data not-usually supported by your database. This data will be encoded by default by pickle, but more control may be added by adding the parameters datatype=... or output_schema=....

Implementations​

Here are a few superduper native implementations:

ObjectModel

Use a self-built model (object) or function with the system:

from superduper import ObjectModel

m = ObjectModel(
    'my-model',
    object=lambda x: x + 2,
)

db.apply(m)

QueryModel

Use a superduper query to extract data from db

from superduper.components.model import QueryModel

query = ... # build a select query
m = QueryModel('my-query', select=query, key='<key-to-extract>')

db.apply(m)

APIModel

Request model outputs hosted behind an API:

from superduper.components.model import APIModel

m = APIModel('my-api', url='http://localhost:6666?token={MY_DEV_TOKEN}&model={model}&text={text}')

db.apply(m)

SequentialModel

Make predictions on the basis of a sequence of models:

from superduper.components.model import SequentialModel

m = SequentialModel(
    'my-sequence',
    models=[
        model1,
        model2,
        model3,
    ]
)

db.apply(m)

See also