Move required path fields up front, persist custom benchmark suites, enlarge readable typography, show full suite benchmark names, add capability-domain trend charts with local Chart.js, and polish ranking name display.
143 lines
4.6 KiB
HTML
143 lines
4.6 KiB
HTML
<!DOCTYPE html>
|
||
<html lang="zh-CN">
|
||
<head>
|
||
<meta charset="UTF-8" />
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||
<title>EvalStone Results</title>
|
||
<link rel="preconnect" href="https://fonts.googleapis.com" />
|
||
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
|
||
<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600;700&display=swap" rel="stylesheet" />
|
||
<link rel="stylesheet" href="/static/styles.css" />
|
||
<script src="/static/vendor/chart.umd.min.js"></script>
|
||
</head>
|
||
<body>
|
||
<div class="page">
|
||
<header class="top">
|
||
<div class="brand">
|
||
<span class="brand-mark"></span>
|
||
<div>
|
||
<h1>EvalStone Results</h1>
|
||
<p>多模型同基准对比 · 扫描 <code>output/</code></p>
|
||
</div>
|
||
</div>
|
||
<nav class="nav-tabs">
|
||
<a href="/">启动台</a>
|
||
<a href="/results" class="active">结果分析</a>
|
||
</nav>
|
||
</header>
|
||
|
||
<main class="results-layout">
|
||
<section class="panel filters-panel">
|
||
<div class="block">
|
||
<h2>数据源</h2>
|
||
<label>
|
||
<span>Output Dir</span>
|
||
<input id="outputDir" />
|
||
</label>
|
||
<div class="actions mt">
|
||
<button type="button" id="reloadBtn" class="primary">重新扫描</button>
|
||
<span id="scanMsg" class="msg"></span>
|
||
</div>
|
||
<p id="scanMeta" class="muted mt"></p>
|
||
</div>
|
||
|
||
<div class="block">
|
||
<h2>选择模型(folder)</h2>
|
||
<div class="bench-actions">
|
||
<button type="button" id="modelAllBtn" class="ghost">全选</button>
|
||
<button type="button" id="modelClearBtn" class="ghost">清空</button>
|
||
</div>
|
||
<div id="modelList" class="pick-list"></div>
|
||
</div>
|
||
|
||
<div class="block">
|
||
<h2>选择 Benchmark</h2>
|
||
<div class="bench-actions">
|
||
<button type="button" id="benchAllBtn" class="ghost">全选</button>
|
||
<button type="button" id="benchClearBtn" class="ghost">清空</button>
|
||
</div>
|
||
<div id="benchList" class="pick-list bench-pick"></div>
|
||
</div>
|
||
|
||
<div class="actions">
|
||
<button type="button" id="compareBtn" class="primary">更新对比</button>
|
||
</div>
|
||
</section>
|
||
|
||
<section class="panel charts-panel">
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>能力分类走势</h2>
|
||
<span class="muted">横轴 = 能力域(数学/代码/智能体等),纵轴 = 该域平均分</span>
|
||
</div>
|
||
<div class="chart-wrap">
|
||
<canvas id="categoryLineChart"></canvas>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>能力分类柱状对比</h2>
|
||
<span class="muted">各模型在同一能力域的平均得分</span>
|
||
</div>
|
||
<div class="chart-wrap">
|
||
<canvas id="categoryBarChart"></canvas>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>得分走势</h2>
|
||
<span class="muted">横轴 = benchmark,纵轴 = score</span>
|
||
</div>
|
||
<div class="chart-wrap">
|
||
<canvas id="lineChart"></canvas>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>柱状对比</h2>
|
||
<span class="muted">同 benchmark 下各模型得分</span>
|
||
</div>
|
||
<div class="chart-wrap">
|
||
<canvas id="barChart"></canvas>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>雷达图</h2>
|
||
<span class="muted">能力轮廓(需 ≥3 个共同 benchmark)</span>
|
||
</div>
|
||
<div class="chart-wrap radar-wrap">
|
||
<canvas id="radarChart"></canvas>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>得分表</h2>
|
||
<span class="muted">平均值(跨 seed / multi-run)</span>
|
||
</div>
|
||
<div class="table-wrap">
|
||
<table id="scoreTable">
|
||
<thead></thead>
|
||
<tbody></tbody>
|
||
</table>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="chart-block">
|
||
<div class="side-head">
|
||
<h2>各 Benchmark 排名</h2>
|
||
</div>
|
||
<div id="rankList" class="rank-list"></div>
|
||
</div>
|
||
</section>
|
||
</main>
|
||
</div>
|
||
<script src="/static/results.js?v=20260729"></script>
|
||
</body>
|
||
</html>
|