pandera.decisions.ProviderCapabilities

class pandera.decisions.ProviderCapabilities(kinds=frozenset({'choice', 'noul', 'score'}), max_options=255, max_levels=10, max_questions=None, reports_confidence=frozenset({'choice', 'score'}), price_per_million_input_tokens=None)

What a provider’s model can be asked, as opposed to how fast.

Decision models differ here even when they share a wire format: a small local model may take fewer options per question than a hosted one, and a classifier-only model may not answer scores at all. The defaults are the question vocabulary’s own bounds – two to 255 options, two to ten levels – which every provider speaking the shared wire format inherits, so a provider only declares where it is narrower.

A provider is checked against these before any request is made, so an unsupported question is an error naming the column rather than a failure from the server on row one.

Attributes

kinds	Question kinds the model answers.
max_levels	Most levels a single Score may have.
max_options	Most options a single Choice may offer.
max_questions	Most questions one request may carry, when the provider caps it.
price_per_million_input_tokens	What input costs, for plan().
reports_confidence	Kinds for which the model reports how certain it is.

Methods

__init__(kinds=frozenset({'choice', 'noul', 'score'}), max_options=255, max_levels=10, max_questions=None, reports_confidence=frozenset({'choice', 'score'}), price_per_million_input_tokens=None)