Laya on ONNX Runtime

ONNX is how most non-Python Laya runtimes work. An exported graph plus the tokenizer and calibration config is enough to answer typed questions with ONNX Runtime, and several projects publish ready-made exports so you can skip the PyTorch conversion step.

76 projects

laya-review

x007007007

Provides local-first code review using Laya decision rules through ONNX Runtime

0·GitHub repo

laya-snapdragon

piffie

Runs Laya typed decisions on Snapdragon X NPUs through ONNX Runtime and Qualcomm QNN

0·GitHub repo

laya-models

dockndevai

Provides quantized ONNX builds of Laya for browser inference

0·GitHub repo

laya-hexagon-npu

EricYu123456

Deploys Laya inference on Qualcomm Hexagon NPU using ONNX Runtime

0·GitHub repo

laya-vs-llms

Ujjwal3115

Benchmarks Laya against hosted language models on developer commit triage and CI safety decisions

0·GitHub repo

vybir

Galactic717

Runs Laya locally with conformal error guarantees and includes score-based game-learning demos

0·GitHub repo

laya

EduardoGHdez

A Ruby library runs the Laya decision model locally with ONNX Runtime

0·GitHub repo

laya-s-dungeon

ValchanOficial

A tactical auto-battler that uses Laya through ONNX Runtime to choose actions each turn

0·GitHub repo

layar

nullean

A .NET port of Laya provides ONNX and TorchSharp inference backends and CLI tools

0·GitHub repo

laya-go

metalagman

Provides an embeddable Go runtime for local Laya inference and typed decision routing

0·GitHub repo

laya_demo

Cokefish9527

Deploys Laya ONNX weights locally on CPU and serves decisions through an HTTP API

0·GitHub repo

laya-node

roryyu

Provides a Node.js SDK for local Laya typed decisions using Transformers.js and ONNX

0·GitHub repo

laya-cpp

samiul000

Implements native C++ Laya inference with ONNX Runtime, benchmarking, and INT8 quantization

0·GitHub repo

vibe-check

ghobs91

Detects ragebait and spam in real time using Laya ONNX

0·GitHub repo

laya-multilingual-qnn

asopitech

Runs Laya's multilingual decision model on Qualcomm Hexagon NPU through ONNX Runtime QNN

0·GitHub repo

jev-systemone-local

katya4oyu

Plans a local Jev-compatible System One server with MLX, Core ML, and ONNX Laya backends

0·GitHub repo

ruby_llm-providers-laya

codenamev

A RubyLLM provider uses local ONNX Laya checkpoints to answer judge questions

0·GitHub repo

ruby_decision_model-providers-laya

codenamev

Adds a local ONNX-backed Laya provider to the Ruby decision-model gem

0·GitHub repo

usejev

ali-master

Serves Laya through native ONNX inference on Bun with a TypeSafe-compatible API and bilingual playground

0·GitHub repo

laya-universal

abusuraihsakhri

Provides cross-platform inference utilities for Laya checkpoints using ONNX Runtime and MLX

0·GitHub repo

laya-cuda-bench

bhushankinge

Benchmarks Laya inference throughput, latency and cost across NVIDIA GPUs and serving backends

0·GitHub repo

xcanv-models

attachemd

Provides opt-in ONNX Laya decision-model packs quantized for local inference

0·GitHub repo

jev-course-demo

clchrf

Evaluates university AI course plans in the browser using quantized Laya ONNX inference

0·GitHub repo

laya-ui

bharath-ui1027

Provides a multilingual Laya decision engine with UI components and optional serving integrations

0·GitHub repo

Use a pre-converted export

inferenceprince/laya-onnx is an fp16 export of the English checkpoint. From its model card:

pip install onnxruntime huggingface_hub tokenizers numpy
hf download inferenceprince/laya-onnx --local-dir laya-onnx

The card includes a full Python example that builds the prompt, runs the graph and applies the fitted temperature. It reports load-to-first-answer of 3 to 5 s against 25 to 35 s for PyTorch on an Intel i5-14400F. The official package also has a laya[onnx] extra with an ONNXAgent class in laya.onnx_agent.

INT8 on CPU

edgejev converts Laya to ONNX, quantizes it to INT8 and serves it:

uv tool install "edgejev[build]"
edgejev build --backend laya --out ./jev-int8

Its README reports 15.6 ms per question for the INT8 multilingual model on a 4 vCPU Xeon (Cascade Lake) and 32.1 ms in fp32, with the model shrinking from 1290 MB to 324 MB. The same table shows AG News accuracy dropping from 92.8% to 91.2%.

ONNX in other languages

Caveats

The graph returns raw logits. Divide by the fitted temperature from rl_agent_config.json before the softmax, or the probabilities mean nothing. Quantized builds drift from PyTorch: the layaForWeb README reports 97.9% top-answer agreement for its int8 build, and it notes that its confidence values are approximate because calibration was fitted on the full-precision model.

More ways to use Laya