Laya – Renesas X5H

Laya: state + typed questions in, calibrated choice / score / yes-no answers out — in a single forward pass

Source: Convai Innovations' Laya model and the receptron/laya ONNX Runtime port

🚧 Coming soon. This model is being optimized and validated for Renesas hardware — no fixed release date yet.

Introduction

Laya is an open-source, multilingual, non-autoregressive System-1 decision model. Given a state (text, email, ticket or JSON) and typed questions, it returns typed answers with calibrated probabilities in a single forward pass — it never generates text, so there is no output to parse and nothing to hallucinate. Renesas is preparing an optimized deployment of this model for the R-Car Gen5 platform, based on the ONNX version.

• Model Architecture: Non-autoregressive System-1 decision / classification model (Jev-compatible).
• Answer types: choice (one option, probability per option), score (expected level on an ordered rubric), noul (calibrated P(true) for a yes/no statement).
• Use cases: Routing, scoring, moderation and guardrails for agentic / LLM pipelines.
• Languages: 100+ (English and multilingual checkpoints).
• Source Model: convaiinnovations/laya (exact checkpoint/version to be confirmed)
• ONNX Reference: receptron/laya — ONNX Runtime implementation, fp32 bundle (~1.7 GB), output matching the Python reference to four decimal places.

This page is a placeholder. Content — including the license and source-model link — is provisional and subject to change before the model is fully published.