CallInfo

Package: flyteplugins.typesafe_ai

What one system_one call cost, plus the calibration a plain field dropped.

Choice, Score and Noul carry their own calibration, so for those the maps below are redundant. They exist for the shorthand field types – a bool or a Literal – which hold a plain value: the calibration is dropped from your model, not from the call, and this is where to find it.

Both maps are dict[str, float] so that a task can return a CallInfo.

Parameters

class CallInfo(
    model: str = '',
    questions: int = 0,
    input_tokens: int = 0,
    output_tokens: int = 0,
    latency_s: float = 0.0,
    values: dict[str, float] = <factory>,
    confidence: dict[str, float] = <factory>,
)