Glossary
Choice question
A choice question asks Laya to pick one answer from a list you define, such as a department, and returns the pick plus a probability for every option.
Swiss-hosted inference. Nothing you send is ever stored. Swiss data residency
Definition
What Choice question means
A choice question is Laya's single-label classification primitive, and one of the three question types of the Jev-style /v1/systemone protocol. You give it instructions and a set of options; it returns exactly one of them.
Shape
criteria can be a dict of label to description, or a plain list of labels (normalised to a dict with no descriptions):
Show technical detailsHide technical details· json sample
The answer contains choice (the argmax label), probabilities (one entry per label, summing to 1), confidence (normalised-entropy confidence, 0 to 1) and an action block.
Designing good choice questions
- Make options mutually exclusive and include an explicit
other. The softmax must put its mass somewhere. - Write short, distinctive descriptions. Options share the
head_max_lenbudget (192 tokens English, 256 multilingual) and each is capped at 48 tokens. - Keep the label count moderate. Laya is strong at small label sets (AG News, 4 labels: 0.950) and weak at large ones at default settings (Banking77, 77 labels: 0.425). For 20+ options, use a two-step coarse-to-fine choice.
- Use a descriptive key rather than a letter unless you are working around noul bias, in which case neutral keys like A/B help.
If more than one label can apply, ask several noul questions instead. See choice, score and noul and the intent detection use case.
Related terms
How Choice question connects to the rest of the vocabulary.
Go deeper
- ExplainerChoice, score and noul: three primitives for typed decisionsChoice picks a label, score places text on a scale, noul gives the probability a statement is true. How each works in Laya, when to use it and where it is weak.
- Use caseIntent detection with a calibrated decision modelDetect what users want in chatbots, voice menus and search. Define intents per request, get calibrated probabilities and hand off to a person when unsure.
Get started
Try it on your own text
Send any text with your own typed questions and get a calibrated probability for every answer. 6k free tokens, no card needed; then $0.0357 per 1M input tokens — 15% below Jev.
Swiss-hosted inference. Nothing you send is ever stored.