User repository on GitHub
Browser-only Laya multilingual demos powered by Transformers.js
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Demos & Experiments
Open Jev Laya
Browser-only Laya demo suite: four Vite apps — a Traditional Chinese decision playground, Gomoku, Big Two and a 3D maze — run the Laya ONNX model on-device through Transformers.js, downloading fp16 weights from Hugging Face on first use.
- Stars
- 11
- Language
- Python
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- killkli
- Last verified
- Oct 2026
- Languages
- PythonTypeScript
- Platforms
- BrowserCloudflare
- Capabilities
- Open sourceRuns locally
Highlights
- Four browser demos — zh-TW decision playground, Gomoku, Big Two, 3D maze — share examples/laya-shared/client.ts.
- OpenJev.load() loads tokenizer, config and ONNX model; decide(state, questions) returns option probabilities.
- First run downloads a ~647 MB fp16 ONNX file; WebGPU is tried first with WebAssembly as fallback.
- Root dev and build commands use the pinned hosted revision 643219b2b3a112ff1c40fb1143f050aa871f76f7.
- Derived from nico-martin/open-jev under MIT; weights live in a separate Hugging Face repository.
Quickstart
bash
pnpm install
pnpm dev:zh-tw-web-layaWatch out
Grounded on the repository README and GitHub metadata captured 2026-10-15; the README documents no accuracy or latency figures for the four demos, so their decision quality cannot be verified from it. The browser downloads weights from Hugging Face at runtime, so first load depends on that service being reachable.
Questions about this entry
Short answers drawn from the linked project or page.
A standalone project that runs the Laya multilingual ONNX model in the browser with Transformers.js. It ships four Vite demos — a Traditional Chinese decision playground, Gomoku, Big Two and a 3D maze — all sharing one client and sequence helper.
No. Inference runs on the user's device and the demos do not call an inference server. On first use the browser downloads the weights from the Hugging Face model repository; the fp16 ONNX file is about 647 MB.
Node.js 24 and pnpm 12 for development. The browser uses WebGPU when available and falls back to WebAssembly; no local Python installation or inference API is needed for these demos.
Run pnpm install, then a dev script such as pnpm dev:zh-tw-web-laya, pnpm dev:gomoku-laya, pnpm dev:big-two-laya or pnpm dev:maze3d-laya. Each demo builds into its own examples/<demo-name>/dist directory.
Yes, optionally: with uv, Python 3.12 and enough disk and memory, pnpm laya:prepare and pnpm laya:quantize export an fp16 ONNX from the upstream model, and the demo then reads local assets from .cache/laya-export/public.
MIT, retaining the license and copyright of nico-martin/open-jev, from which the browser demos and the typed-decision compatibility layer are derived. The Laya model comes from its upstream Hugging Face repository.
OpenJev.load() loads the tokenizer, configuration and model, then decide(state, questions) evaluates a set of typed questions in one model forward pass and returns probability distributions over the supplied options.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Text-to-screenshot (Malay Vasa)
X: Google's MediaPipe Decision Maker, tested in the browser
X: Playwright CLI + Jev vs Playwright MCP
X: jev-ultrafast splits every browser decision into operation, target and value. JEV passed 18,000 stars in its…
More Jev resources like this
A playable Three.js driving simulator where Jev-powered autopilot picks steering and speed up to four times per second: compact tables of eligible paths, traffic and road boundaries go to Jev while geometry and control math stay local.
Jev Demos & Experiments#community#catalog-import#cobanov
Community
standardagents/jevpilotJavaScript
standardagents/jevpilotJavaScriptA local proof of concept where Jev picks which of two conflicting signal phases gets the green light at a real Hong Kong intersection, using near real-time keyless open-data traffic sensors. A phase-timing state machine enforces minimum and maximum green, amber and all-red around the model's answer.
Jev Demos & Experiments#community#catalog-import#logicrw
The Cua computer-use platform documents a jev-use recipe where TypeSafe Jev picks bounded action IDs and Cua Driver executes clicks and fills in the browser.
Jev Demos & Experiments#community#browser#computer-use
Community
trycua/cuaRust
trycua/cuaRustAn agent skill that reproduces Jev-style single-step logits over indexed browser controls using local Qwen weights for offline browser use.
Jev Demos & Experiments#community#browser#open-models
Community
APUS-AI-Lab/fast-browser-usePython
APUS-AI-Lab/fast-browser-usePythonA small local app that turns Jev's per-pixel probability distributions into paintings: stroke pigments are sampled from the probabilities, mean color supplies the underpainting, and entropy controls relief, rendered in a browser worker.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
achimala/jevinciJavaScript
achimala/jevinciJavaScriptA public operator lab for TypeSafe Jev: a chess match as the flagship page, a row of smaller judgment booths, and one shared client that keeps the API key out of the browser via a local Python server or a Cloudflare Worker.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
4anti/jev-broadcast-labJavaScript
4anti/jev-broadcast-labJavaScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.