Laya
by ConvAI Innovations ~ Routing, classification, and guardrail decisions with calibrated probabilities
Laya is ConvAI Innovations' non-autoregressive decision model. It does not generate text: it reads a state and answers typed questions in a single forward pass, each with calibrated probabilities. A `noul` question returns the probability that a statement is true, a `choice` question picks one of named options, and a `score` question places the state on a scale. It is designed for triage, routing, and guardrails, and answers many questions about the same input in one request.
`convaiinnovations/laya` (alias `laya`) sends each request to its English checkpoint, a 421M-parameter ModernBERT-large encoder with a decision head, or to its multilingual checkpoint for 100+ languages. Call it at `POST /v1/systemone`.
Call it
Send a state and named questions to POST /v1/systemone. Each question is a noul, choice or score question, and each answer comes back with its probabilities. This model does not accept chat completions.
curl https://api.inference.boundless.network/v1/systemone \
-H "Authorization: Bearer $BOUNDLESS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "convaiinnovations/laya",
"state": "Please refund my duplicate payment.",
"questions": {
"refund": {
"type": "noul",
"instructions": "Does the customer request a refund?"
}
}
}'