Spaces:
Running
Running
File size: 19,345 Bytes
256345e 9c40bae 256345e 9c40bae 256345e 9c40bae 256345e 9c40bae 256345e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Teachable LLM - In-Browser Transfer Learning w/ Transformers.js</title>
<meta name="description"
content="Interactive demonstration of in-browser text classification using Transformers.js feature extraction and k-Nearest Neighbors with customizable model selection." />
<!-- Google Fonts: Google Sans, Google Sans Text, Roboto, Roboto Mono -->
<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=Google+Sans:wght@400;500;700&family=Google+Sans+Text:wght@400;500;700&family=Roboto:wght@400;500;700&family=Roboto+Mono:wght@400;500&display=swap"
rel="stylesheet">
<link rel="stylesheet" href="style.css" />
<!-- Importmap for @xenova/transformers browser module resolution -->
<script type="importmap">
{
"imports": {
"@xenova/transformers": "https://cdn.jsdelivr.net/npm/@xenova/transformers@2.17.2"
}
}
</script>
</head>
<body>
<!-- Top Google Four-Color Accent Bar -->
<div class="google-color-bar">
<div class="bar-blue"></div>
<div class="bar-red"></div>
<div class="bar-yellow"></div>
<div class="bar-green"></div>
</div>
<div class="app-container">
<!-- Header -->
<header class="app-header">
<div class="brand">
<div class="brand-icon">
<svg width="28" height="28" viewBox="0 0 24 24" fill="none">
<path
d="M12 2C6.48 2 2 6.48 2 12s4.48 10 10 10 10-4.48 10-10S17.52 2 12 2zm0 18c-4.41 0-8-3.59-8-8s3.59-8 8-8 8 3.59 8 8-3.59 8-8 8z"
fill="#4285F4" />
<circle cx="12" cy="8" r="2.5" fill="#EA4335" />
<circle cx="8" cy="14" r="2.5" fill="#FBBC04" />
<circle cx="16" cy="14" r="2.5" fill="#34A853" />
</svg>
</div>
<div>
<h1 class="brand-title">Teachable LLM</h1>
<p class="brand-subtitle">Transfer Learning via Transformers.js & k-NN in WebAssembly</p>
</div>
</div>
<!-- Model Selector & Status Controls -->
<div class="header-controls">
<div class="model-selector-card">
<label for="model-select">Transformer Model:</label>
<select id="model-select" class="model-select">
<option value="Xenova/all-MiniLM-L6-v2" selected>all-MiniLM-L6-v2 (384-d, ~23MB - Default)</option>
<option value="Xenova/all-mpnet-base-v2">all-mpnet-base-v2 (768-d, ~110MB - High Accuracy)</option>
<option value="Xenova/bge-small-en-v1.5">bge-small-en-v1.5 (384-d, ~67MB - BAAI BGE)</option>
<option value="Xenova/gte-small">gte-small (384-d, ~67MB - General Embeddings)</option>
<option value="Xenova/distilbert-base-uncased">distilbert-base-uncased (768-d, ~67MB - DistilBERT)</option>
</select>
</div>
<!-- Model Load Status Badge -->
<div class="status-card">
<div class="status-indicator">
<span id="status-dot" class="status-dot loading"></span>
<span id="status-text">Initializing pipeline...</span>
</div>
<div id="progress-container" class="progress-container" style="display: none;">
<div id="progress-bar" class="progress-bar" style="width: 0%;"></div>
</div>
</div>
</div>
</header>
<!-- Concept Overview Hero & Pipeline Stepper Card -->
<section class="hero-card">
<div class="hero-content">
<div class="hero-header-row">
<div>
<h2>Interactive Architecture Pipeline</h2>
<p class="hero-subtitle">Explore how Transformers convert text into high-dimensional vectors and perform
transfer learning in-browser.</p>
</div>
<button type="button" id="animate-pipeline-btn" class="btn btn-secondary btn-sm">
<svg width="14" height="14" viewBox="0 0 24 24" fill="none" style="margin-right:4px;">
<polygon points="5,3 19,12 5,21" fill="currentColor" />
</svg>
Animate Forward Pass
</button>
</div>
<div class="stepper-grid">
<div class="stepper-step active" data-step="1">
<div class="step-num step-blue">1</div>
<div class="step-details">
<h3>Tokenization</h3>
<p>Subword splitting & position IDs</p>
</div>
</div>
<div class="stepper-step" data-step="2">
<div class="step-num step-red">2</div>
<div class="step-details">
<h3>Self-Attention</h3>
<p>Multi-Head token interaction</p>
</div>
</div>
<div class="stepper-step" data-step="3">
<div class="step-num step-yellow">3</div>
<div class="step-details">
<h3>Mean Pooling</h3>
<p>Dense vector fingerprint</p>
</div>
</div>
<div class="stepper-step" data-step="4">
<div class="step-num step-green">4</div>
<div class="step-details">
<h3>2D Space & k-NN</h3>
<p>PCA Manifold & Class Regions</p>
</div>
</div>
</div>
</div>
</section>
<!-- Main Content Grid -->
<main class="main-grid">
<!-- Left Column: Interactive Controls -->
<div class="left-column">
<!-- Training Dataset Management Card -->
<section class="panel-card">
<div class="panel-header">
<h2>1. Training Dataset</h2>
<span id="dataset-count" class="badge-neutral">0 examples</span>
</div>
<!-- Add Example Form -->
<form id="add-example-form" class="add-form">
<div class="form-group">
<label for="example-text">Text Example</label>
<input type="text" id="example-text" placeholder="e.g. The service was fast and super friendly!"
required />
</div>
<div class="form-group">
<label for="example-label">Class Label</label>
<div class="label-input-group">
<input type="text" id="example-label" placeholder="e.g. positive" required />
<div class="preset-badges">
<button type="button" class="preset-badge badge-positive"
data-label="positive">positive</button>
<button type="button" class="preset-badge badge-negative"
data-label="negative">negative</button>
<button type="button" class="preset-badge badge-neutral"
data-label="neutral">neutral</button>
</div>
</div>
</div>
<button type="submit" class="btn btn-secondary">+ Add Example</button>
</form>
<!-- Dataset Example List -->
<div class="dataset-section">
<div class="dataset-actions">
<h3>Current Memory Dataset</h3>
<div class="btn-group">
<button type="button" id="reset-demo-btn" class="btn btn-sm btn-outline">Reset Demo
Dataset</button>
<button type="button" id="clear-dataset-btn" class="btn btn-sm btn-outline-danger">Clear
All</button>
</div>
</div>
<div id="dataset-list" class="dataset-list">
<!-- Populated dynamically by demo.js -->
</div>
</div>
</section>
<!-- Inference & Prediction Card -->
<section class="panel-card">
<div class="panel-header">
<h2>2. Run Inference</h2>
</div>
<div class="inference-form">
<div class="form-group">
<label for="test-text">Test Input Sentence</label>
<textarea id="test-text" rows="2"
placeholder="Type any sentence to test classification..."></textarea>
</div>
<div class="form-row">
<div class="form-group inline">
<label for="k-value">Neighbors (k):</label>
<input type="number" id="k-value" value="3" min="1" max="10" />
</div>
<button type="button" id="predict-btn" class="btn btn-primary">Classify Text</button>
</div>
</div>
<!-- Prediction Result Display -->
<div id="result-box" class="result-box" style="display: none;">
<div class="result-header">
<div class="result-main">
<span class="result-title">Predicted Class:</span>
<div id="predicted-label"></div>
</div>
<div class="confidence-box">
<span class="confidence-label">Confidence: <strong id="confidence-val">--</strong></span>
<div class="confidence-track">
<div id="confidence-fill" class="confidence-fill" style="width: 0%;"></div>
</div>
</div>
</div>
<div id="vector-preview" class="vector-preview"></div>
<!-- Top k Neighbors Breakdown -->
<div class="neighbors-section">
<h4>Nearest Neighbor Votes</h4>
<div id="neighbors-list" class="neighbors-list"></div>
</div>
</div>
</section>
</div>
<!-- Right Column: Architecture & Logs -->
<div class="right-column">
<!-- Activity Log Card -->
<section class="panel-card log-panel">
<div class="panel-header">
<h2>Activity Console</h2>
<span class="live-tag">LIVE</span>
</div>
<div id="log-output" class="log-output">
<!-- Streamed console output -->
</div>
</section>
<!-- Architecture Summary Card -->
<section class="panel-card info-card">
<div class="panel-header">
<h2>Architecture Specs</h2>
</div>
<ul class="specs-list">
<li><strong>Active Model:</strong> <code id="spec-model-name">Xenova/all-MiniLM-L6-v2</code></li>
<li><strong>Runtime:</strong> ONNX Runtime Web via Transformers.js</li>
<li><strong>Embedding Dimension:</strong> <span id="spec-embed-dim">384</span></li>
<li><strong>Pooling Strategy:</strong> Mean pooling with L2 normalization</li>
<li><strong>Classifier:</strong> k-Nearest Neighbors (Euclidean Distance)</li>
</ul>
</section>
</div>
</main>
<!-- Full-Width Interactive Architecture & Storytelling Visualizer Section -->
<section class="viz-storytelling-section">
<div class="section-title-row">
<div>
<h2>Deep Learning Visual Explorers</h2>
<p>Interactive step-by-step breakdown inspired by Transformer Explainer & GAN Lab</p>
</div>
</div>
<div class="viz-grid">
<!-- Viz 1: Subword Tokenizer -->
<div class="panel-card viz-card" id="viz-step-1">
<div class="panel-header">
<h2>Stage 1: Subword Tokenization</h2>
<span class="badge label-custom">WordPiece / BPE</span>
</div>
<p class="viz-desc">Input text is mapped into discrete subword token IDs and positional indices.</p>
<div id="tokenizer-viz-container" class="viz-container"></div>
</div>
<!-- Viz 2: Self-Attention Matrix -->
<div class="panel-card viz-card" id="viz-step-2">
<div class="panel-header">
<h2>Stage 2: Self-Attention Matrix & QKV Arcs</h2>
<span class="badge badge-negative">Softmax(QK<sup>T</sup>/√d)V</span>
</div>
<p class="viz-desc">Tokens attend to each other across Multi-Head Self-Attention layers to capture
contextual meaning.</p>
<div id="attn-arcs-container" class="attn-arcs-wrapper"></div>
<div id="attention-viz-container" class="viz-container"></div>
</div>
<!-- Viz 3: Pooling & Embedding Fingerprint Barcode -->
<div class="panel-card viz-card" id="viz-step-3">
<div class="panel-header">
<h2>Stage 3: Sentence Embedding Fingerprint</h2>
<span class="badge badge-neutral">Mean Pooling & L2 Norm</span>
</div>
<p class="viz-desc">Token vectors are collapsed into a dense 1D fingerprint representing the entire
sentence's semantics.</p>
<div id="pooling-viz-container" class="viz-container"></div>
</div>
<!-- Viz 4: 2D Embedding Space & Decision Boundary Canvas (GAN Lab style) -->
<div class="panel-card viz-card full-width-viz" id="viz-step-4">
<div class="panel-header">
<div>
<h2>Stage 4: 2D Embedding Space & k-NN Decision Boundaries</h2>
<p class="viz-desc">High-dimensional embeddings projected to 2D via PCA. Background shows memory
class zones; click canvas to query!</p>
</div>
<div id="pca-legend" class="pca-legend"></div>
</div>
<div class="canvas-wrapper">
<canvas id="pca-canvas" width="760" height="420"></canvas>
<div class="canvas-hint-overlay">
<span>💡 Click anywhere on canvas to test a 2D vector coordinate query</span>
</div>
</div>
</div>
</div>
</section>
<footer class="app-footer">
<p>Built with <a href="https://github.com/xenova/transformers.js" target="_blank" rel="noopener">Transformers.js</a>
</p>
</footer>
</div>
<!-- Main Module Script -->
<script type="module" src="demo.js"></script>
</body>
</html> |