TokenVisualizer

Running

File size: 8,793 Bytes

<!doctype html>
<html lang="en">
<head>
  <meta charset="utf-8" />
  <meta name="viewport" content="width=device-width, initial-scale=1" />
  <title>TokenVisualizer — Minimal</title>
  <link rel="preconnect" href="https://fonts.googleapis.com">
  <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;600&family=JetBrains+Mono:wght@400;600&display=swap" rel="stylesheet">
  <style>
    :root{
      --bg:#0b0f14; --text:#ffffff; --muted:#9aa4b2; --accent:#38bdf8; --border:#1f2a3a;
      --card1:#0c1624; --card2:#0a1220; --chip:#111827; --chip-border:#263246; --chip-hover:#1a2434;
      --mono:'JetBrains Mono',ui-monospace,Menlo,Consolas,monospace; --sans:Inter,system-ui,-apple-system,"Segoe UI",Roboto,Ubuntu,"Helvetica Neue",Arial;
    }
    *{box-sizing:border-box} body{margin:0;background:radial-gradient(900px 500px at 10% -10%, #07314a, transparent),var(--bg);color:var(--text);font-family:var(--sans)}
    .container{max-width:1100px;margin:0 auto;padding:1.25rem}
    header{padding-top:1.5rem} h1{margin:.2rem 0 .4rem;font-size:1.9rem}
    .sub{color:var(--muted);margin:.25rem 0 1rem}
    .card{background:linear-gradient(180deg,var(--card1),var(--card2));border:1px solid var(--border);border-radius:14px;padding:1rem;box-shadow:0 10px 40px rgba(0,0,0,.35)}
    label span{color:var(--muted);font-size:.9rem}
    select,textarea{width:100%;border-radius:10px;border:1px solid var(--border);background:#0a1220;color:var(--text);padding:.7rem .85rem;outline:none}
    select:focus,textarea:focus{border-color:var(--accent)}
    .controls{display:grid;gap:.8rem;margin-bottom:1rem}
    .row{display:flex;gap:.75rem;align-items:center}
    .status{color:var(--muted)}
    .grid{display:grid;gap:1rem;grid-template-columns:1fr}
    @media (min-width:900px){.grid{grid-template-columns:1fr 1fr}}
    .head{display:flex;align-items:center;justify-content:space-between;margin-bottom:.5rem}
    .tokens{display:flex;flex-wrap:wrap;gap:.5rem;max-height:360px;overflow:auto;padding:.25rem}
    .chip{border:1px solid var(--chip-border);background:var(--chip);padding:.35rem .5rem;border-radius:10px;font-family:var(--mono);font-size:.9rem;transition:background .12s,border-color .12s}
    .chip:hover{background:var(--chip-hover);border-color:var(--accent)}
    .chip.active{outline:2px solid var(--accent)}
    pre.ids{font-family:var(--mono);background:#0a1220;border:1px solid var(--border);border-radius:10px;padding:.75rem;max-height:360px;overflow:auto;white-space:pre-wrap}
    .caption{color:var(--muted);font-size:.9rem;margin-top:.5rem}
    footer{color:var(--muted);text-align:center;padding:1.25rem 0 2rem}
    a{color:var(--accent)}
  </style>
</head>
<body>
  <header class="container">
    <h1>TokenVisualizer</h1>
    <p class="sub">Live view of tokens and token IDs. Powered by Transformers.js — all in your browser.</p>
  </header>

  <main class="container">
    <section class="card controls">
      <label>
        <span>Model</span>
        <select id="model">
          <!-- Tip: keep this first so the demo works instantly once you upload /assets/gpt2/* -->
          <option value="local:gpt2">GPT-2 (local, fast)</option>
          <option value="Xenova/gpt2">GPT-2 (Hub)</option>
          <option value="Xenova/llama2-tokenizer">Llama-2 (Hub)</option>
          <option value="Xenova/mistral-tokenizer">Mistral (Hub)</option>
          <option value="Xenova/gemma-tokenizer">Gemma (Hub)</option>
          <option value="Xenova/bert-base-uncased">BERT Base Uncased (Hub)</option>
        </select>
      </label>
      <label>
        <span>Text</span>
        <textarea id="input" rows="3">Hello world! This is a tokenizer demo.</textarea>
      </label>
      <div class="row">
        <span id="status" class="status">Loading tokenizer…</span>
      </div>
    </section>

    <section class="grid">
      <article class="card">
        <div class="head"><h3>Tokens</h3></div>
        <div id="tokens" class="tokens"></div>
        <p class="caption">Tokens are subword chunks the model learned from lots of text.</p>
      </article>

      <article class="card">
        <div class="head"><h3>Token IDs</h3></div>
        <pre id="ids" class="ids"></pre>
        <p class="caption">IDs are how the model “sees” tokens — just numbers.</p>
      </article>
    </section>
  </main>

  <footer class="container">
    <small>Built by Peter Adams • Powered by <a href="https://github.com/xenova/transformers.js" target="_blank" rel="noreferrer">Transformers.js</a></small>
  </footer>

  <!-- Minimal, robust script (no copy/export) -->
  <script type="module">
    // Prefer keeping all requests on huggingface.co to avoid CORS/VPN issues.
    // Option 1 (simple): CDN import (works on many networks)
    const tf = await import('https://cdn.jsdelivr.net/npm/@xenova/[email protected]');
    // Option 2 (bulletproof): self-host the file in your Space and use:
    // const tf = await import('./assets/vendor/transformers.min.js');

    tf.env.useBrowserCache = true;
    tf.env.allowLocalModels = true; // <-- REQUIRED for local folder paths
    tf.env.localModelPath = '/';

    const $ = s => document.querySelector(s);
    const modelSel = $('#model');
    const inputEl  = $('#input');
    const statusEl = $('#status');
    const tokensEl = $('#tokens');
    const idsEl    = $('#ids');

    // Single state object; never reassign
    const state = { tokens: [], ids: [] };
    let tokenizer = null;
    let runId = 0;

    const status = (msg) => { statusEl.textContent = msg; };
    const debounce = (fn, ms=200) => { let t; return (...a)=>{ clearTimeout(t); t=setTimeout(()=>fn(...a), ms); }; };

    async function loadTokenizer(modelId){
      status('Loading tokenizer…');
      try {
        if (modelId === 'local:gpt2') {
          // Note: no double slashes, no /resolve/main – just your folder.
          tokenizer = await tf.AutoTokenizer.from_pretrained('assets/gpt2');
        } else {
          tokenizer = await tf.AutoTokenizer.from_pretrained(modelId);
        }
        status('Tokenizer ready.');
      } catch (e) {
        console.error('Tokenizer load failed:', e);
        tokenizer = null;
        status('Failed to load tokenizer (network blocked or slow). Try GPT-2 or a different VPN route.');
      }
    }


    async function tokenize(){
      const myRun = ++runId;
      if (!tokenizer) {
        await loadTokenizer(modelSel.value);
        if (!tokenizer) { render(); return; } // keep UI stable on failure
      }

      const text = (inputEl.value ?? '').trim();
      if (!text) {
        state.tokens = [];
        state.ids = [];
        render();
        status('Type to tokenize…');
        return;
      }

      status('Tokenizing…');
      try {
        const enc = await tokenizer.encode(text);
        if (myRun !== runId) return; // drop stale results
        state.tokens = Array.isArray(enc?.tokens) ? enc.tokens : [];
        state.ids    = Array.isArray(enc?.ids)    ? enc.ids    : [];
        render();
        status(`Done. ${state.tokens.length} tokens.`);
      } catch (e) {
        console.error('Tokenize failed:', e);
        state.tokens = state.tokens ?? [];
        state.ids    = state.ids ?? [];
        render();
        status('Error tokenizing. See console.');
      }
    }

    function render(){
      const tokens = Array.isArray(state.tokens) ? state.tokens : [];
      const ids    = Array.isArray(state.ids)    ? state.ids    : [];

      // Tokens pane
      tokensEl.innerHTML = '';
      tokens.forEach((tok, i) => {
        const chip = document.createElement('span');
        chip.className = 'chip';
        chip.dataset.i = i;
        chip.textContent = tok;
        chip.addEventListener('mouseenter', ()=>highlight(i, true));
        chip.addEventListener('mouseleave', ()=>highlight(i, false));
        tokensEl.appendChild(chip);
      });

      // IDs pane
      idsEl.textContent = ids.join(' ');

      if (tokens.length === 0) status('Type to tokenize…');
    }

    function highlight(i, on){
      const ids = Array.isArray(state.ids) ? state.ids : [];
      if (!ids.length) return;

      const parts = ids.map((id, idx) => (idx === i && on) ? `[${id}]` : String(id));
      idsEl.textContent = parts.join(' ');

      const chip = tokensEl.querySelector(`[data-i="${i}"]`);
      if (chip) chip.classList.toggle('active', on);
    }

    const debounced = debounce(tokenize, 200);
    inputEl.addEventListener('input', debounced);

    modelSel.addEventListener('change', async ()=>{
      tokenizer = null;                 // force reload
      await loadTokenizer(modelSel.value);
      tokenize();
    });

    // Initial load
    await loadTokenizer(modelSel.value);
    tokenize();
  </script>
</body>
</html>