Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
★ 35,583⑂ 3,192👁 123⑃ 1,210● 221▤ 67
+42
Daily
+320
Weekly
+3,045
Monthly
35,583
stars
3,192
forks
123
watchers
1,210
prs
221
issues
67
releases
README
Cite this work. M, N.; M, A. From Laya to Vega: Open Non-Autoregressive Decision Models with Calibrated Probabilities. Preprints 2026, 2026110781. https://doi.org/10.20944/preprints202611.0781.v1
BibTeX
bibtex
@article{m2026laya_vega,
title = {From Laya to Vega: Open Non-Autoregressive Decision Models with Calibrated Probabilities},
author = {M, N. and M, A.},
journal = {Preprints},
year = {2026},
pages = {2026110781},
doi = {10.20944/preprints202611.0781.v1},
url = {https://doi.org/10.20944/preprints202611.0781.v1}
}
Multilingual, non-autoregressive System 1 decision engine. Typed decisions over 100+ languages in a single forward pass — 33 ms — trained with reinforcement learning against strictly proper scoring rules (RLCD), with a router that picks the right checkpoint per request.
Train your own decision engine at studio.embedding.in. Studio, from Embedding Space, trains a System 1 decision engine from a description of your company: it writes the questions and the examples, fits and calibrates a vega model, and gives you a signed policy file that runs on your own machine with vegaml. No labelled data is needed to start, and a new workspace gets $10 of free credit.
Installation
bash
python -m pip install laya
With uv, run uv add laya in a uv project or uv pip install laya in a virtual environment.
Python 3.10 or newer. Optional extras: laya[serve] (HTTP server), laya[mcp] (MCP server), laya[langchain] (LangChain and LangGraph), laya[llamaindex] (LlamaIndex selectors), laya[crewai] (CrewAI routing), laya[onnx] (ONNX Runtime), laya[fast] (TileLang GPU fast path), laya[structured] (pydantic models in decide), laya[vision] (Pillow and torchvision, for predict(image=...) and predict(video=...)), laya[audio] (soundfile, to decode audio files). Step-by-step setup for each platform, CPU-only or GPU PyTorch builds, and troubleshooting are in Installation details.
For TypeScript / Node.js / browser, see laya-ts/. npm releases (npm install laya-ts) are published from this repository's laya-ts-v release tags.
Long documents. laya-multilingual reads up to 8,192 tokens with max_len=8192. Measured accuracy and time by document length, reproducible with research/scripts/bench_long_context.py:
Quickstart
One argument puts vegaml's engine under laya's names: a frozen 0.8B reader with a small physics engine on top, so every answer carries a conformal prediction set, an abstain flag, an out-of-distribution flag and a route decided without a threshold you had to guess at. Text, images, speech and 73k-token documents go through the same call. Python 3.10 or newer; details in vega inside laya.
bash
python -m pip install -U laya 0.4.23 or newer: vegaml comes with it
python
import laya
agent = laya.Agent(vegaml=True) the 0.8B universal reader (nandakishorm/vega-U), the default
Switching the model underneath is one string. Every option keeps the same engine, questions and fields:
laya.ATop issues
#390Docs site: structure and the guides still to write21 comments · 2026-09-24#102Implemented the coarse-to-fine shortlist your docs recommend: BANKING77 54.3% → 60.8%, latency halved17 comments · 2026-09-21#131laya-multilingual never selects the first-listed score option — in English too (0/290), while laya does14 comments · 2026-09-22#919[feature]: serve your own checkpoints beside the built-ins, routed by what the request asks12 comments · 2026-10-04#208The committed 51-language sweep's ECE and confidence columns predate the #42 temperature clamp and no longer reproduce9 comments · 2026-09-23#7Feature Request: TypeScript / Node.js / Bun.js Support10 comments · 2026-09-19#635[feature]: Stability under option reorder or rename as an opt-in reliability signal, alongside min_confidence8 comments · 2026-09-27#285Re-benchmark the checkpoints and refresh README and BENCHMARKS.md on each release8 comments · 2026-09-23
Discussions
#1500Evaluating multi-question decisions: per-question gains versus whole-case reliability0 comments · 2026-11-15#1497Evaluating yes/no/unknown: refuted conditions vs insufficient evidence, separately from abstention0 comments · 2026-11-13#1492Analysis: Training a System One model using Laya, and how it compares to Jev0 comments · 2026-11-11#1490Multiple Option ✅0 comments · 2026-11-09#1298Integration report: fine-tuned Laya as a LangGraph router, vs XGBoost, BERT and Gemini1 comments · 2026-10-20#1270Licensing question: training data composition + commercial deployment of laya-multilingual1 comments · 2026-10-17#1198Recommended settings for CPU-only training with limited RAM0 comments · 2026-10-13#1048Laya dotnet NuGet1 comments · 2026-10-08
Latest release
v0.4.34 v0.4.34 2026-11-15Full Changelog: https://github.com/NandhaKishorM/laya/compare/v0.4.33...v0.4.34
相关报道 (6)
News & coverage (8)
Hacker NewsLaya: Multilingual, non-autoregressive System 1 decision engine▲ 2 · 0 comments · 2026-10-01startupfortuneJEV, LAYA and CLEF - System One Models Bring a New AI Architecture2026-10-03startupfortuneOpen Source Is Pulling Away From JEV as Nandakishor Mukkunnoth's LAYA Hits 121 Contributors2026-10-01noteMaking local Laya the 'decision-maker' for ChatGPT: From Mac auto-start to operation2026-09-29Cloudflare BlogIntroducing Clef: Our open-source decision models, and new RL fine-tuning platform2026-10-01Pasquale PillitteriStanford and NVIDIA release CLM-8B, 9 times faster than Jev2026-09-28Startup FortuneThe Developer Who Says He Built Jev First Just Gave Everyone a Free Version2026-09-24KuCoinThe Laya-MLX model can run on 1GB of local memory, enabling fast decision-making.2026-09-22