Laya
Non-autoregressive System 1 decision engine — typed decisions in a single forward pass, 100+ languages
The open implementation of the viral Jev model: a non-autoregressive System 1 decision engine making typed choice, score, and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router. 20K stars in its first week.
Open foundation models — the base layers everything builds on.
Key features
- Non-autoregressive decision engine returning typed decisions in one forward pass
- 33 ms decisions across 100+ languages with a per-request checkpoint router
- Typed outputs: choice, ordered rubric score, or probability of a proposition
- Trained with reinforcement learning against proper scoring rules (RLCD)
- Extras for HTTP serving, MCP, LangChain, LlamaIndex, CrewAI, and ONNX
Good for
- Routing support tickets by department, urgency, and churn risk
- Low-latency classification and scoring inside agent workflows
Stack & tags
PythonTypeScriptdecision-engineclassificationmultilingualnon-autoregressiverlpipopen-weights
Quick start
python -m pip install layaFrom the project README on GitHub.