runs/17.html

<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>gemma-4-12b-it-heretic – Local LLM Benchmarks</title>
<link rel="stylesheet" href="../style.css">
</head>
<body>
<header class="top">
  <a class="brand" href="../index.html">Local LLM Benchmarks</a>
  <nav>
    <a href="../index.html">Overview</a>
    <a href="../methodology.html">Methodology</a>
  </nav>
</header>
<main>
<p class="sub"><a href="../index.html#v3-quick">← Overview</a></p>
<h1>gemma-4-12b-it-heretic</h1>
<dl class="meta">
  <dt>Quantization</dt><dd>NVFP4</dd>
  <dt>Reasoning</dt><dd>off</dd>
  <dt>Profile</dt><dd>quick</dd>
  <dt>Suite</dt><dd>v2b</dd>
  <dt>Date</dt><dd>2026-09-18</dd>
  <dt>Hardware</dt><dd>NVIDIA DGX Spark, 128 GB RAM</dd>
  <dt>Inference software</dt><dd>vLLM 0.26.0</dd>
  <dt>Measured</dt><dd>directly at the inference server</dd>
  <dt>Overall</dt><dd class="strong">57.7</dd>
</dl>

<h2>Categories</h2>
<div class="table-wrap">
<table>
<thead><tr><th>Category</th><th class="num">Score</th></tr></thead>
<tbody>
<tr><td>Knowledge &amp; reasoning</td><td class="num">63.7</td></tr>
<tr><td>Math</td><td class="num">48.4</td></tr>
<tr><td>Instruction following</td><td class="num">46.8</td></tr>
<tr><td>Code</td><td class="num">62.0</td></tr>
<tr><td>German</td><td class="num">67.7</td></tr>
</tbody>
</table>
</div>

<h2>Tasks</h2>
<div class="table-wrap">
<table>
<thead><tr><th>Category</th><th>Task</th><th class="num">Score</th><th class="num">±</th><th class="num">n</th></tr></thead>
<tbody>
<tr>
  <td>Knowledge &amp; reasoning</td>
  <td>mmlu_pro</td>
  <td class="num">76.7</td>
  <td class="num">± 2.4</td>
  <td class="num">300</td>
</tr>
<tr>
  <td>Knowledge &amp; reasoning</td>
  <td>gpqa_diamond_gen</td>
  <td class="num">56.1</td>
  <td class="num">± 3.5</td>
  <td class="num">198</td>
</tr>
<tr>
  <td>Knowledge &amp; reasoning</td>
  <td>musr_gen</td>
  <td class="num">58.4</td>
  <td class="num">± 3.1</td>
  <td class="num">250</td>
</tr>
<tr>
  <td>Math</td>
  <td>omni_math_gen</td>
  <td class="num">48.4</td>
  <td class="num">± 3.2</td>
  <td class="num">250</td>
</tr>
<tr>
  <td>Instruction following</td>
  <td>ifbench</td>
  <td class="num">33.0</td>
  <td class="num">± 2.7</td>
  <td class="num">300</td>
</tr>
<tr>
  <td>Instruction following</td>
  <td>ifeval</td>
  <td class="num">88.7</td>
  <td class="num">± 2.6</td>
  <td class="num">150</td>
</tr>
<tr>
  <td>Code</td>
  <td>livecodebench_gen</td>
  <td class="num">62.0</td>
  <td class="num">± 3.4</td>
  <td class="num">200</td>
</tr>
<tr>
  <td>German</td>
  <td>mmlu_prox_de</td>
  <td class="num">70.0</td>
  <td class="num">± 2.7</td>
  <td class="num">300</td>
</tr>
<tr>
  <td>German</td>
  <td>include_de_gen</td>
  <td class="num">63.0</td>
  <td class="num">± 3.9</td>
  <td class="num">154</td>
</tr>
</tbody>
</table>
</div>
<p class="hint">± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.</p>

<h2>Speed</h2>
<div class="table-wrap">
<table>
<thead><tr><th>Measure</th><th class="num">Context (target)</th><th class="num">Context (actual)</th><th class="num">Median</th><th class="num">p10</th><th class="num">p90</th><th class="num">n</th></tr></thead>
<tbody>
<tr>
  <td>Generation [t/s]</td>
  <td class="num">–</td>
  <td class="num">–</td>
  <td class="num">21.4</td>
  <td class="num">21.4</td>
  <td class="num">21.4</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Prompt processing [t/s]</td>
  <td class="num">512</td>
  <td class="num">528</td>
  <td class="num">5070.5</td>
  <td class="num">4947.7</td>
  <td class="num">5118.3</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Prompt processing [t/s]</td>
  <td class="num">4096</td>
  <td class="num">4014</td>
  <td class="num">5258.0</td>
  <td class="num">5221.7</td>
  <td class="num">5293.9</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Prompt processing [t/s]</td>
  <td class="num">16384</td>
  <td class="num">15985</td>
  <td class="num">3514.1</td>
  <td class="num">3505.2</td>
  <td class="num">3518.5</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Time to first token [ms]</td>
  <td class="num">512</td>
  <td class="num">528</td>
  <td class="num">103.9</td>
  <td class="num">103.2</td>
  <td class="num">106.7</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Time to first token [ms]</td>
  <td class="num">4096</td>
  <td class="num">4014</td>
  <td class="num">763.4</td>
  <td class="num">758.2</td>
  <td class="num">768.7</td>
  <td class="num">5</td>
</tr>
<tr>
  <td>Time to first token [ms]</td>
  <td class="num">16384</td>
  <td class="num">15985</td>
  <td class="num">4548.8</td>
  <td class="num">4543.1</td>
  <td class="num">4560.4</td>
  <td class="num">5</td>
</tr>
</tbody>
</table>
</div>
<p class="hint">Measured directly at the inference server.</p>
<footer class="page-foot">As of 2026-10-10 · scores 0–100 · built with <code>bench publish</code></footer>
</main>
<script src="../chart.umd.min.js"></script>
<script src="../app.js"></script>
</body>
</html>