Laya

An Android app that ranks your options entirely on-device with a quantized Laya via ONNX Runtime, with no network needed.

Ports the laya-mlx Snake demo to the DGX Spark GB10, showing the open Laya decision model working on Nvidia hardware.

An OpenAI-compatible Go gateway where Jev or a local Laya picks which LLM answers each prompt, with a decision benchmark.

A Rust server with custom Metal kernels that runs Laya in 15.5 ms on an M5 Mac, up to 4.1x faster than the reference server, behind the same API.

Mgear runs the receptron/laya-onnx weights in Unity through onnxruntime-unity and plays tic-tac-toe against the model.

Measures the decision head's attention drop-off on long inputs across 400 documents, then chunks, screens and aggregates, or locates the one passage that matters.

Reads text or an ebook and rates it on 82 calibrated dimensions such as genre, mood, themes and style, at about a second per page on an AMD Strix Halo.

Improves candidate recall in the lime Chinese input method for Squirrel and adds experimental local Laya MLX reranking based on context.

A Snake game whose moves all come from a fine-tuned copy of Laya, showing a decision model acting in a real-time control loop.

Adds optional on-device Laya routing to OpenZL lossless compression, choosing a compressor adaptively for integer data on Apple silicon with a size guard.

A modern Sudoku game where the Laya decision model picks moves with about 35 ms of GPU latency.

Gets Laya running a Snake demo offline on a RISC-V board, with a reproducible Bianbu setup, ONNX adaptation and tests.

Two decision models fight a real Doom deathmatch with the same state, questions and shield: open local Laya against hosted Jev.

Cross-platform Agent Skill plus MCP tool that gives coding agents a local Laya decision capability. Verified on Apple Silicon with MLX and PyTorch backends.

Ahmed made the 322M multilingual checkpoint drive every paddle move in Breakout, locally on a Windows laptop with zero cloud. Code in Eng-Ahmd/laya-breakout.

Uses Laya on a CPU-only Cloud Run container to pick the right Gemini tier for each ADK request and scores the setup by cost per successful task.

Daniel F built a fighting game with 21 world-leader-inspired fighters, signature moves, combos and replays. The opponent can be driven by Laya in the browser.

Dima Nurm sorted 1,000 emails with the 322M Laya model on an M3 Air: 28.6 s without screen recording, 65.1% accuracy against reference labels, 0 MB swap, no cloud.

The decision model never writes code; each step it picks the next tool, scores progress, estimates risk and says if the goal is reached. Any OpenAI-compatible LLM executes.

Fine-tunes Laya to label GitHub issues and only applies labels when the model is confident, leaving the uncertain ones for people.

Routes support tickets by intent, scores urgency, frustration and churn risk, and escalates by confidence, all with a local System 1 model.

A GitHub Action that triages issues with a local Laya model, needing no API key or external service.

A duel bot for Ultima Online that lets local Laya and cloud Jev fight, with a live monitor comparing their decisions.

Combines Laya, LangGraph, Qwen 3 8B on Ollama and an MCP sandbox into a local coding assistant with a kid-friendly UI.