Spaces:

pufanyi
/

vbvrpro_output

Running

App Files Files Community

vbvrpro_output / index.html

pufanyi

Add shared step-0 SFT ODE origin to score trajectories

900e9f6 verified 15 days ago

Raw

History Blame Contribute Delete

7.48 kB

	<!doctype html>
	<html lang="en">
	<head>
	<meta charset="utf-8" />
	<meta name="viewport" content="width=device-width, initial-scale=1" />
	<meta
	name="description"
	content="Explore VBVR-Pro task scores, baseline deltas, and every scored video across the strict In-Domain DanceGRPO sweep."
	/>
	<meta name="theme-color" content="#f3f0e8" />
	<title>VBVR-Pro Output Explorer</title>
	<link rel="stylesheet" href="./styles.css" />
	</head>
	<body>
	<div class="page-shell">
	<header class="hero">
	<nav class="topbar" aria-label="Primary">
	<a class="brand" href="./" aria-label="VBVR-Pro Output Explorer home">
	<span class="brand-mark" aria-hidden="true">V</span>
	<span>VBVR·PRO / OUTPUT ATLAS</span>
	</a>
	<div class="topbar-meta">
	<span class="status-dot" aria-hidden="true"></span>
	<span id="dataset-status">Loading evaluation archive…</span>
	</div>
	</nav>

	<div class="hero-grid">
	<div>
	<p class="eyebrow">WAN2.2 TI2V-5B · DANCEGRPO</p>
	<h1>See where training moved the needle.</h1>
	<p class="hero-copy">
	Task-level change, sample-level evidence, and every scored video from the
	strict In-Domain sweep—always grounded against the same SFT baseline.
	</p>
	</div>
	<div class="hero-stamp" aria-label="Dataset summary">
	<div class="stamp-number" id="hero-run-count">20</div>
	<div class="stamp-label">evaluated runs</div>
	<div class="stamp-rule"></div>
	<div class="stamp-detail">
	<span id="hero-video-count">10,500</span> videos ·
	<span id="hero-task-count">100</span> tasks
	</div>
	</div>
	</div>
	</header>

	<main>
	<section class="control-deck" aria-labelledby="run-selection-title">
	<div class="section-heading compact">
	<div>
	<p class="section-index">01 / RUN SELECTION</p>
	<h2 id="run-selection-title">Choose a checkpoint and sampler</h2>
	</div>
	<div class="baseline-pill">
	<span>BASELINE</span>
	SFT epoch 1
	</div>
	</div>

	<div class="run-controls">
	<label class="select-field">
	<span>Checkpoint</span>
	<select id="checkpoint-select" aria-label="Checkpoint"></select>
	</label>
	<fieldset class="mode-fieldset">
	<legend>Sampling mode</legend>
	<div id="mode-switcher" class="mode-switcher"></div>
	</fieldset>
	</div>

	<div id="score-cards" class="score-cards" aria-live="polite"></div>
	</section>

	<section class="trend-section" aria-labelledby="trend-title">
	<div class="section-heading">
	<div>
	<p class="section-index">02 / SCORE TRAJECTORY</p>
	<h2 id="trend-title">Overall score across training</h2>
	</div>
	<p class="section-note">Step 0 is the shared SFT ODE baseline; all runs use the same 500 samples.</p>
	</div>
	<div class="chart-card">
	<div id="trend-legend" class="trend-legend" aria-label="Chart legend"></div>
	<div id="trend-chart" class="trend-chart" role="img" aria-label="Overall score trend chart"></div>
	</div>
	</section>

	<section class="task-section" aria-labelledby="task-title">
	<div class="section-heading">
	<div>
	<p class="section-index">03 / TASK LEDGER</p>
	<h2 id="task-title">Every task, ranked by change</h2>
	</div>
	<p id="task-summary" class="section-note"></p>
	</div>

	<div class="task-toolbar">
	<label class="search-field">
	<span class="sr-only">Search tasks</span>
	<svg aria-hidden="true" viewBox="0 0 24 24">
	<path d="m21 21-4.35-4.35m2.35-5.65a8 8 0 1 1-16 0 8 8 0 0 1 16 0Z" />
	</svg>
	<input id="task-search" type="search" placeholder="Search exact task_name…" autocomplete="off" />
	</label>

	<div id="domain-filter" class="domain-filter" aria-label="Domain filter">
	<button class="active" type="button" data-domain="all">All</button>
	<button type="button" data-domain="In_Domain">In-Domain</button>
	<button type="button" data-domain="Out_of_Domain">Out-of-Domain</button>
	</div>

	<label class="sort-field">
	<span>Sort</span>
	<select id="task-sort">
	<option value="delta-desc">Largest gain</option>
	<option value="delta-asc">Largest drop</option>
	<option value="score-desc">Highest score</option>
	<option value="score-asc">Lowest score</option>
	<option value="name-asc">Task name</option>
	</select>
	</label>
	</div>

	<div class="task-table-wrap">
	<table class="task-table">
	<thead>
	<tr>
	<th scope="col">Task</th>
	<th scope="col">Domain / category</th>
	<th scope="col">Current</th>
	<th scope="col">Baseline</th>
	<th scope="col">Δ vs baseline</th>
	<th scope="col"><span class="sr-only">Inspect</span></th>
	</tr>
	</thead>
	<tbody id="task-table-body"></tbody>
	</table>
	<div id="empty-state" class="empty-state" hidden>
	No tasks match this filter.
	</div>
	</div>
	</section>
	</main>

	<footer>
	<span>VBVR-PRO · MAIN_V2</span>
	<span>MANIFEST 326F7BDA · EVALKIT 42A1593D</span>
	</footer>
	</div>

	<dialog id="task-dialog" class="task-dialog">
	<div class="dialog-shell">
	<header class="dialog-header">
	<div>
	<p id="dialog-kicker" class="section-index"></p>
	<h2 id="dialog-title"></h2>
	<p id="dialog-raw-name" class="raw-task-name"></p>
	</div>
	<button id="dialog-close" class="close-button" type="button" aria-label="Close task detail">×</button>
	</header>

	<div class="dialog-body">
	<div id="dialog-metrics" class="dialog-metrics"></div>
	<details class="prompt-panel">
	<summary>Evaluation prompt</summary>
	<p id="dialog-prompt"></p>
	</details>

	<div class="sample-heading">
	<div>
	<p class="section-index">SAMPLE EVIDENCE</p>
	<h3>Selected run vs SFT baseline</h3>
	</div>
	<p>Five matched samples. Scores are exact values from the scorer JSON.</p>
	</div>
	<div id="sample-grid" class="sample-grid"></div>
	</div>
	</div>
	</dialog>

	<div id="loading-overlay" class="loading-overlay" role="status">
	<div class="loading-card">
	<div class="loading-mark">V</div>
	<p>Indexing the evaluation archive</p>
	<div class="loading-line"><span></span></div>
	</div>
	</div>

	<script src="./app.js" defer></script>
	</body>
	</html>