238 lines
8.7 KiB
HTML
238 lines
8.7 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 Launch</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=Inter:wght@400;500;600;700&display=swap" rel="stylesheet" />
|
||
<link rel="stylesheet" href="/static/styles.css" />
|
||
</head>
|
||
<body>
|
||
<div class="page">
|
||
<header class="top">
|
||
<div class="brand">
|
||
<span class="brand-mark"></span>
|
||
<div>
|
||
<h1>EvalStone Launch</h1>
|
||
<p>基于 <code>bash/run.py</code> 的评测启动台</p>
|
||
</div>
|
||
</div>
|
||
<div class="top-right">
|
||
<nav class="nav-tabs">
|
||
<a href="/" class="active">启动台</a>
|
||
<a href="/results">结果分析</a>
|
||
</nav>
|
||
<div class="health">
|
||
<span id="healthDot" class="dot"></span>
|
||
<span id="healthText">connecting…</span>
|
||
</div>
|
||
</div>
|
||
</header>
|
||
|
||
<!-- Top action bar -->
|
||
<section class="action-bar">
|
||
<label class="field-grow">
|
||
<span>Model</span>
|
||
<input id="model" placeholder="DeepSeek-V4-Flash-Int8" />
|
||
</label>
|
||
<label class="field-grow">
|
||
<span>API URL</span>
|
||
<input id="api_url" placeholder="http://localhost:30000/v1" />
|
||
</label>
|
||
<label>
|
||
<span>API Key</span>
|
||
<input id="api_key" type="password" placeholder="EMPTY" />
|
||
</label>
|
||
<label class="switch-inline">
|
||
<input type="checkbox" id="thinking" />
|
||
<span class="switch-slider"></span>
|
||
<span class="switch-label">Thinking</span>
|
||
</label>
|
||
<div class="action-btns">
|
||
<button type="button" id="launchBtn" class="primary">▶ 启动</button>
|
||
<button type="button" id="stopBtn" class="danger" disabled>■ 停止</button>
|
||
</div>
|
||
</section>
|
||
|
||
<main class="layout">
|
||
<!-- Left: configuration tabs -->
|
||
<section class="panel main-panel">
|
||
<div class="tabs" role="tablist">
|
||
<button type="button" class="tab active" data-tab="scope">评测范围</button>
|
||
<button type="button" class="tab" data-tab="config">路径配置</button>
|
||
<button type="button" class="tab" data-tab="advanced">高级参数</button>
|
||
</div>
|
||
|
||
<!-- Scope tab -->
|
||
<div id="scopeTab" class="tab-pane active">
|
||
<div class="mode-bar">
|
||
<span class="mode-label">选择方式</span>
|
||
<div class="mode-toggle">
|
||
<button type="button" class="mode-btn active" data-mode="suite">按 Suite</button>
|
||
<button type="button" class="mode-btn" data-mode="datasets">自选 Benchmark</button>
|
||
</div>
|
||
<span id="scopeEstimate" class="estimate"></span>
|
||
</div>
|
||
|
||
<div id="suitePane">
|
||
<div class="section-title">内置 Suite</div>
|
||
<div id="suiteList" class="suite-grid"></div>
|
||
|
||
<div class="section-title flex-between">
|
||
<span>自定义组合</span>
|
||
<span id="customSuiteHint" class="hint">在「自选」模式勾选并保存</span>
|
||
</div>
|
||
<div id="customSuiteList" class="suite-grid"></div>
|
||
</div>
|
||
|
||
<div id="datasetsPane" class="hidden">
|
||
<div class="bench-toolbar">
|
||
<input id="benchSearch" class="search" placeholder="搜索 benchmark…" />
|
||
<button type="button" id="selectAllBtn" class="ghost">全选</button>
|
||
<button type="button" id="clearAllBtn" class="ghost">清空</button>
|
||
<button type="button" id="saveCustomBtn" class="ghost">保存组合</button>
|
||
<span id="selectedCount" class="counter">已选 0</span>
|
||
</div>
|
||
<div id="categoryFilters" class="category-chips"></div>
|
||
<div id="benchmarkList" class="benchmark-grid"></div>
|
||
</div>
|
||
</div>
|
||
|
||
<!-- Config tab -->
|
||
<div id="configTab" class="tab-pane">
|
||
<div class="path-grid">
|
||
<label class="full">
|
||
<span>Dataset Dir <small class="muted">（evalscope 会自动在其下找 datasets/）</small></span>
|
||
<input id="dataset_dir" />
|
||
</label>
|
||
<label class="full">
|
||
<span>Output Dir</span>
|
||
<input id="output_dir" />
|
||
</label>
|
||
<label class="full">
|
||
<span>Config YAML</span>
|
||
<input id="config" />
|
||
</label>
|
||
<label class="full">
|
||
<span>Tokenizer Path</span>
|
||
<input id="tokenizer_path" />
|
||
</label>
|
||
</div>
|
||
<div class="preset-bar">
|
||
<span class="muted">路径预设</span>
|
||
<button type="button" class="ghost" id="presetDefaultBtn">默认</button>
|
||
<button type="button" class="ghost" id="presetCurrentBtn">保存当前</button>
|
||
<button type="button" class="ghost" id="presetLoadBtn">加载预设</button>
|
||
</div>
|
||
</div>
|
||
|
||
<!-- Advanced tab -->
|
||
<div id="advancedTab" class="tab-pane">
|
||
<div class="grid-4">
|
||
<label>
|
||
<span>Folder Name</span>
|
||
<input id="folder_name" placeholder="默认 model / model_THINKING" />
|
||
</label>
|
||
<label>
|
||
<span>Limit</span>
|
||
<input id="limit" placeholder="none" />
|
||
</label>
|
||
<label>
|
||
<span>Seed</span>
|
||
<input type="number" id="seed" value="42" />
|
||
</label>
|
||
<label>
|
||
<span>Batch Size</span>
|
||
<input type="number" id="batch_size" value="4" min="1" />
|
||
</label>
|
||
<label>
|
||
<span>max_tokens_add</span>
|
||
<input type="number" id="max_tokens_add" value="0" min="0" step="1024" />
|
||
</label>
|
||
<label>
|
||
<span>thinking scale</span>
|
||
<input type="number" id="thinking_max_tokens_scale" value="1.0" min="0.1" step="0.1" />
|
||
</label>
|
||
<label>
|
||
<span>Write Summary</span>
|
||
<select id="write_summary">
|
||
<option value="true" selected>是</option>
|
||
<option value="false">否</option>
|
||
</select>
|
||
</label>
|
||
</div>
|
||
<details class="fold">
|
||
<summary>Judge 模型配置</summary>
|
||
<div class="grid-2 mt">
|
||
<label>
|
||
<span>Judge Model</span>
|
||
<input id="judge_model" />
|
||
</label>
|
||
<label>
|
||
<span>Judge API URL</span>
|
||
<input id="judge_api_url" />
|
||
</label>
|
||
<label>
|
||
<span>Judge API Key</span>
|
||
<input id="judge_api_key" type="password" />
|
||
</label>
|
||
<label>
|
||
<span>Judge Max Tokens</span>
|
||
<input type="number" id="judge_max_tokens" />
|
||
</label>
|
||
</div>
|
||
</details>
|
||
</div>
|
||
|
||
<div id="formMsg" class="form-msg"></div>
|
||
</section>
|
||
|
||
<!-- Right: status panel -->
|
||
<section class="panel status-panel">
|
||
<div class="active-head">
|
||
<h2>运行状态</h2>
|
||
<span id="activeStatus" class="badge idle">idle</span>
|
||
</div>
|
||
<div id="activeCard" class="active-card idle">
|
||
<div class="active-id"><code id="activeJobId">—</code></div>
|
||
<div class="progress-wrap">
|
||
<div id="progressBar" class="progress-bar" style="width:0%"></div>
|
||
</div>
|
||
<div id="progressText" class="progress-text"></div>
|
||
</div>
|
||
|
||
<div class="cmd-wrap">
|
||
<div class="cmd-head">
|
||
<span>命令预览</span>
|
||
<button type="button" id="copyCmdBtn" class="ghost" disabled>复制</button>
|
||
</div>
|
||
<pre id="activeCmd" class="cmd">尚未启动任务</pre>
|
||
</div>
|
||
|
||
<div class="log-wrap">
|
||
<div class="log-head">
|
||
<span>实时日志</span>
|
||
<label class="check-inline">
|
||
<input type="checkbox" id="autoScroll" checked />
|
||
自动滚动
|
||
</label>
|
||
</div>
|
||
<pre id="logView" class="log-view"></pre>
|
||
</div>
|
||
|
||
<div class="jobs-wrap">
|
||
<div class="jobs-head">
|
||
<h3>历史任务</h3>
|
||
<button type="button" id="refreshJobsBtn" class="ghost">刷新</button>
|
||
</div>
|
||
<div id="jobList" class="job-list"></div>
|
||
</div>
|
||
</section>
|
||
</main>
|
||
</div>
|
||
<script src="/static/app.js?v=2"></script>
|
||
</body>
|
||
</html>
|