English → AlgoLisp

A trained T5-small model translates plain English into validated AlgoLisp entirely in your browser. Compare the native WebGPU compute path with the WebGL2 fragment-shader fallback below.

60.5M BF16 parameters No server inference No JS ML framework autolisp-rs validation

1. Common English examples

Click one to copy it
Preferred backend

2. WebGPU demo

Skip to WebGL2 ↓

Compute pipelines on a dedicated, surface-free WebGPU device. The first load downloads and verifies about 122 MB of content-addressed weights.

WebGPU inference is ready to load.

Loading is deferred until this panel approaches the viewport.

Compatibility backend

3. WebGL2 demo

Examples ↑

The same model expressed as bounded fullscreen fragment passes over floating-point and integer textures. Previously fetched weights can be reused from the browser cache.

WebGL2 inference is ready to load.

This creates a separate fallback renderer and model instance.