JavaScript

jsMain and webMain use @litertjs/core to load and run models in the browser. The current build uses npm package version 2.6.0.

Model and runtime

Put the model in your application's web static resources, serve it via a web server, and pass an accessible URL such as /models/CelsiusToFahrenheit.tflite to the inference example. The URL must match your deployment base path; cross-origin loading also requires server permission.

LiteRTCompiler.init() waits for shared web runtime initialization. WASM assets load from a CDN by default; offline or restricted environments can configure a self-hosted path before initialization starts.

Acceleration and limitations

CPU uses WASM; GPU requests select WASM when WebGPU is unavailable; NPU requests use WebNN options. Browser, device, model, and server environment affect results. The current implementation does not use signatureIndex to select a signature.

Writing inputs creates JS tensors; outputs are populated after run(). The repository contains browser CPU tests. Their /base/... paths belong to the test server, not application deployments.