Spaces:
Running
title: Laya WebGPU Quick Start
emoji: 🦊
colorFrom: indigo
colorTo: blue
sdk: static
app_file: index.html
pinned: false
Laya WebGPU Quick Start
Try urgency, sentiment, and refund-triage decisions entirely in your browser. Choose the English, multilingual, or typed-decisions checkpoint, then run it on WebGPU (FP32) or WASM/CPU (mixed weight-only INT8). Model files are fetched from immutable Hugging Face revisions; messages and inference stay on your device.
Browser GPU support varies. If WebGPU is unavailable, choose WASM/CPU. Initial model downloads are large, and browser acceptance for full-model WebGPU remains in progress. This Space does not contain an API or server-side inference.
Run locally with your JEV API key
The Space repository also includes a self-contained Docker build under
local-docker/. Clone this Space repository, then run from its root:
git clone https://huggingface.co/spaces/ti3x-m/laya-webgpu
cd laya-webgpu
docker build -t laya-quickstart-jev ./local-docker
docker run --rm -p 127.0.0.1:7860:7860 laya-quickstart-jev
Open http://127.0.0.1:7860, select JEV API · local development, and enter
your own key. Docker builds that local variant from the bundled source and runs
a same-origin proxy to JEV. The key is held in the browser page, sent to JEV
through your local container, and not saved by this app. The published Space
remains static and does not expose the JEV option or proxy.
Models: English, multilingual, and typed decisions.