bootstrap_few_shot¶

bootstrap_few_shot(
    fn,
    data,
    *,
    teacher=None,
    expected=None,
    metric=None,
    max_bootstrapped=4,
    max_labeled=16,
    num_threads=8,
    seed=0,
)

An improved copy: the function (or a stronger teacher model) runs on rows with known answers, and the runs that were right become worked examples, whole (reasoning and tool calls included); labeled rows fill the rest.

Examples¶

import functai
from functai import *

# not run: the teacher answers every row (Make it better runs one)
taught = functai.bootstrap_few_shot(team, train, teacher="gpt-6-sol", expected="category")