Transformers.js

<script type="module">
import { pipeline } from 'https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.0/+esm';
let start = performance.now();
const embed = await pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2',
  { device: 'webgpu', dtype: 'fp32' });                  // ONNX Runtime Web, WebGPU provider
const load = (performance.now() - start) / 1000;
const titles = ['The Quiet Harbor', 'Patterns of the Deep Web', 'Salt and Saffron',
  'Small Steps to Big Summits', "The Clockmaker's Paradox", 'Gardens in Glass'];
start = performance.now();
const vectors = await embed(['a cookbook of spicy recipes', ...titles],
  { pooling: 'mean', normalize: true });                 // one unit vector per text
const [query, ...books] = vectors.tolist();
const run = performance.now() - start;
const cosine = (a, b) => a.reduce((sum, x, i) => sum + x * b[i], 0);
const ranked = titles.map((t, i) => [cosine(query, books[i]), t]).sort((a, b) => b[0] - a[0]);
console.log(`load ${load.toFixed(1)} s, 7 texts ${run.toFixed(0)} ms, ${query.length} dims`);
for (const [score, title] of ranked.slice(0, 3)) console.log(score.toFixed(3), title);
window.__done = true;
</script>

load 28.7 s, 7 texts 434 ms, 384 dims
0.267 Salt and Saffron
0.154 Patterns of the Deep Web
0.145 Gardens in Glass