functai for TypeScript - v0.1.0
    Preparing search index...

    Interface InstructionSearchOptions<F, R>

    interface InstructionSearchOptions<F = AnyAIFunction, R = Rec> {
        candidates?: number;
        concurrency?: number;
        expected?: Expected<F, R>;
        finalists?: number;
        maxBootstrapped?: number;
        maxLabeled?: number;
        metric?: Metric<R>;
        minibatch?: number;
        promptLm?: string;
        seed?: number;
        teacher?: string;
        threshold?: number;
        trials?: number;
        valset?: readonly R[];
    }

    Type Parameters

    Hierarchy (View Summary)

    Index
    candidates? concurrency? expected? finalists? maxBootstrapped? maxLabeled? metric? minibatch? promptLm? seed? teacher? threshold? trials? valset?
    candidates?: number

    Instructions (the current one and proposals) and demo sets to try (default 6).

    concurrency?: number

    Runs in flight at once (default 4).

    expected?: Expected<F, R>

    Where the right answers are, as in evaluate.

    finalists?: number

    How many top combinations are scored on every validation row (default 3).

    maxBootstrapped?: number
    maxLabeled?: number
    metric?: Metric<R>

    Which runs are good: (row, outputs) => score; default exact_match against the row's answers.

    minibatch?: number

    Rows per minibatch (default 20).

    promptLm?: string

    The model that writes instructions (default: the configured one).

    seed?: number
    teacher?: string

    A stronger model to write the examples ("gpt-4.1"); default the function's own.

    threshold?: number

    A run counts when its score is at least this (default: any score above 0).

    trials?: number

    Minibatch evaluations (default 12).

    valset?: readonly R[]

    Rows to score on (default: the rows).

    Settings
    Member Visibility

    On This Page
    Properties