src/embeddings.js

import { config } from './config.js';

const sleep = (ms) => new Promise(r => setTimeout(r, ms));

export async function embed(texts, kind = 'passage') {
  const inputs = (Array.isArray(texts) ? texts : [texts]).map(t => `${kind}: ${t}`);
  const url = `${config.hfApiUrl}/hf-inference/models/${config.embedModel}/pipeline/feature-extraction`;
  // Retry: o endpoint hf-inference às vezes dá 503/504 (modelo frio) — tenta de novo.
  let lastErr = '';
  for (let attempt = 1; attempt <= 4; attempt++) {
    try {
      const res = await fetch(url, {
        method: 'POST',
        headers: { Authorization: `Bearer ${config.hfToken}`, 'Content-Type': 'application/json' },
        body: JSON.stringify({ inputs, options: { wait_for_model: true } }),
      });
      if (res.ok) return res.json();
      lastErr = `HF embed ${res.status}`;
    } catch (e) {
      lastErr = `HF embed: ${e.message}`;
    }
    if (attempt < 4) await sleep(attempt * 1500); // backoff: 1.5s, 3s, 4.5s
  }
  throw new Error(`${lastErr} (após 4 tentativas)`);
}

export function cosine(a, b) {
  let dot = 0, na = 0, nb = 0;
  for (let i = 0; i < a.length; i++) { dot += a[i]*b[i]; na += a[i]*a[i]; nb += b[i]*b[i]; }
  return dot / (Math.sqrt(na) * Math.sqrt(nb) + 1e-8);
}