LabeledFewShot¶
LabeledFewShot(k=16, *, sample=True, seed=0)
Rows with known answers become worked examples, sent before every call.
The cheapest optimizer: no model is called to optimize. Use it as
fn.opt(rows, optimizer=LabeledFewShot(k=8)), or by name:
functai.labeled_few_shot(fn, rows, k=8). One AI function only (a
@module needs BootstrapFewShot, which knows which step each
example belongs to).
Parameters¶
| Name | Type | Description | Default |
|---|---|---|---|
| k | int | At most this many examples (default 16). | 16 |
| sample | bool | True (default): a random sample of the rows; False: the first k. |
True |
| seed | int | The sample's seed, so the same rows give the same examples. | 0 |