Spaces:
Running
Running
File size: 13,827 Bytes
1df72c3 edf774b 1df72c3 edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 1df72c3 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 1df72c3 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd 32a93fb edf774b 32a93fb 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd 1df72c3 edf774b 1df72c3 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 4e141cd edf774b 32a93fb edf774b 32a93fb edf774b 32a93fb edf774b 1df72c3 32a93fb 1df72c3 4e141cd |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 |
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Small But Mighty: Analyzing Compact Math Models</title>
<style>
body {
font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, 'Open Sans', 'Helvetica Neue', sans-serif;
line-height: 1.6;
color: #333;
max-width: 1000px;
margin: 0 auto;
padding: 20px;
background-color: #f9f9f9;
}
h1, h2, h3, h4 {
color: #2d3748;
}
h1 {
font-size: 2.5rem;
margin-bottom: 2rem;
text-align: center;
color: #1a365d;
border-bottom: 3px solid #4299e1;
padding-bottom: 0.5rem;
}
h2 {
font-size: 1.8rem;
margin-top: 2rem;
border-bottom: 1px solid #e2e8f0;
padding-bottom: 0.5rem;
}
.container {
background-color: white;
border-radius: 8px;
padding: 20px;
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
margin-bottom: 20px;
}
.model-card {
background-color: #f0f5ff;
padding: 15px;
border-radius: 8px;
margin-bottom: 15px;
border-left: 4px solid #4299e1;
}
.model-info {
display: flex;
justify-content: space-between;
margin-bottom: 10px;
}
.model-name {
font-weight: bold;
font-size: 1.2rem;
color: #2b6cb0;
}
.model-params {
background-color: #ebf8ff;
padding: 3px 8px;
border-radius: 12px;
font-size: 0.9rem;
color: #2c5282;
}
.performance-highlight {
background-color: #e6fffa;
padding: 10px;
border-radius: 4px;
margin-top: 10px;
border-left: 3px solid #38b2ac;
}
.chart-container {
display: flex;
justify-content: space-between;
flex-wrap: wrap;
gap: 20px;
margin: 30px 0;
}
.chart {
flex: 1;
min-width: 300px;
background: white;
padding: 15px;
border-radius: 8px;
box-shadow: 0 2px 4px rgba(0, 0, 0, 0.05);
}
/* Updated image styling for zoom capability */
.chart img {
width: 100%;
height: auto;
border-radius: 4px;
cursor: zoom-in;
transition: transform 0.3s ease;
}
/* Modal styles for zoomed image */
.modal {
display: none;
position: fixed;
z-index: 1000;
left: 0;
top: 0;
width: 100%;
height: 100%;
overflow: auto;
background-color: rgba(0,0,0,0.9);
}
.modal-content {
margin: auto;
display: block;
max-width: 90%;
max-height: 90%;
}
.close {
position: absolute;
top: 15px;
right: 35px;
color: #f1f1f1;
font-size: 40px;
font-weight: bold;
transition: 0.3s;
cursor: pointer;
}
.close:hover,
.close:focus {
color: #bbb;
text-decoration: none;
}
.chart-caption {
text-align: center;
margin-top: 10px;
font-style: italic;
color: #718096;
}
table {
width: 100%;
border-collapse: collapse;
margin: 20px 0;
}
th, td {
padding: 12px 15px;
text-align: left;
border-bottom: 1px solid #e2e8f0;
}
th {
background-color: #edf2f7;
font-weight: bold;
}
tr:hover {
background-color: #f7fafc;
}
.highlight {
background-color: #fffbeb;
}
.footer {
margin-top: 40px;
text-align: center;
color: #718096;
font-size: 0.9rem;
}
.key-finding {
background-color: #ebf8ff;
padding: 15px;
border-radius: 8px;
margin: 20px 0;
border-left: 4px solid #3182ce;
}
.key-finding h3 {
margin-top: 0;
color: #2b6cb0;
}
</style>
</head>
<body>
<div class="container">
<h1>Small But Mighty: Compact Mathematical Reasoning Models</h1>
<div class="key-finding">
<h3>Key Insight</h3>
<p>The data shows that extremely small language models (under 200M parameters) can demonstrate measurable mathematical abilities, challenging the assumption that large parameter counts are required for basic math reasoning.</p>
</div>
<h2>Model Overview</h2>
<p>This analysis compares the performance of various language models on mathematical problems, with a focus on very compact models:</p>
<div class="model-card">
<div class="model-info">
<span class="model-name">PingVortex/VLM-1</span>
<span class="model-params">124M parameters</span>
</div>
<p>The highest performer among tested models, demonstrating noteworthy mathematical abilities despite its tiny parameter count. Shows particular strength in division operations with 16.7% accuracy.</p>
<div class="performance-highlight">
<strong>Overall math accuracy:</strong> 6.9% on 1000 test questions
</div>
</div>
<div class="model-card">
<div class="model-info">
<span class="model-name">FlameF0X/MathGPT2</span>
<span class="model-params">81.9M parameters</span>
</div>
<p>The second best performer, scoring 2.2% overall accuracy. Shows more balanced performance across addition, subtraction and multiplication operations.</p>
<div class="performance-highlight">
<strong>Operation strength:</strong> 5.4% accuracy on multiplication
</div>
</div>
<div class="model-card">
<div class="model-info">
<span class="model-name">aquiffoo/aquif-r1-1b</span>
<span class="model-params">1.14B parameters + CoT reasoning</span>
</div>
<p>Despite being significantly larger and incorporating chain-of-thought reasoning capabilities, this model shows no measurable performance on the tested mathematical problems.</p>
</div>
<h2>Performance Analysis</h2>
<div class="chart-container">
<div class="chart">
<img src="8d6951f7-dafc-44b4-bd85-c213d2ae77a6.png" alt="Chart showing model accuracy by operation type" class="zoomable">
<div class="chart-caption">Figure 1: Accuracy by Operation Type (%)</div>
</div>
<div class="chart">
<img src="e7692aaa-7623-4e6f-b9f2-23ef0531ebe0.png" alt="Chart showing model performance on math problems" class="zoomable">
<div class="chart-caption">Figure 2: Correct vs Incorrect Answers (1000 questions each)</div>
</div>
</div>
<!-- Modal for image zoom -->
<div id="imageModal" class="modal">
<span class="close">×</span>
<img class="modal-content" id="zoomedImage">
</div>
<h2>Detailed Accuracy Results</h2>
<table>
<thead>
<tr>
<th>Model</th>
<th>Addition (+)</th>
<th>Subtraction (-)</th>
<th>Multiplication (*)</th>
<th>Division (/)</th>
<th>Exponentiation (**)</th>
<th>Overall</th>
</tr>
</thead>
<tbody>
<tr class="highlight">
<td>VLM-1 (124M)</td>
<td>2.6%</td>
<td>5.7%</td>
<td>5.4%</td>
<td>16.7%</td>
<td>0.0%</td>
<td>6.9%</td>
</tr>
<tr>
<td>MathGPT2 (81.9M)</td>
<td>0.8%</td>
<td>2.1%</td>
<td>5.4%</td>
<td>0.9%</td>
<td>0.0%</td>
<td>2.2%</td>
</tr>
<tr>
<td>aquif-r1-1b (1.14B)</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
</tr>
<tr>
<td>BrainrotLM2-Assistant-362M</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
</tr>
<tr>
<td>gonzalez-v1</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
</tr>
<tr>
<td>Snowflake-G0-Release</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
<td>0.0%</td>
</tr>
</tbody>
</table>
<h2>Key Observations</h2>
<ul>
<li><strong>Size doesn't always matter:</strong> VLM-1 with only 124M parameters outperforms larger models on mathematical tasks.</li>
<li><strong>Operation specialization:</strong> VLM-1 shows particularly strong performance on division operations (16.7%).</li>
<li><strong>Architectural importance:</strong> The results suggest that architecture design and training approach may be more important than raw parameter count for specialized tasks.</li>
<li><strong>Zero performance:</strong> Four of the tested models showed no measurable mathematical ability on this test set.</li>
<li><strong>Chain-of-thought limitations:</strong> Despite having CoT capabilities, aquif-r1-1b did not demonstrate mathematical reasoning abilities in this benchmark.</li>
</ul>
<div class="key-finding">
<h3>Research Implications</h3>
<p>These results challenge conventional wisdom about language model scaling laws for mathematical reasoning capabilities. They suggest that highly specialized, compact models might offer a more efficient approach for specific reasoning tasks compared to general-purpose large models.</p>
</div>
<h2>Conclusion</h2>
<p>This analysis demonstrates that extremely small language models can exhibit measurable mathematical reasoning abilities, with models as small as 81.9M and 124M parameters showing the ability to solve basic arithmetic problems. The standout performer, VLM-1 with only 124M parameters, achieved nearly 7% accuracy on a diverse set of 1000 mathematical questions.</p>
<p>These findings suggest that efficient architectural design and specialized training approaches may be more important than raw parameter count when optimizing for specific reasoning capabilities. This could have significant implications for resource-constrained applications where deploying massive models is impractical.</p>
<p>Future research directions could include investigating what specific architectural choices enable these compact models to perform mathematical operations, and how these insights might be applied to develop more efficient specialized models for other reasoning tasks.</p>
<div class="footer">
<p>Data analysis based on benchmark results for VLM-1 (124M), MathGPT2 (81.9M), aquif-r1-1b (1.14B), and other models</p>
<p>© 2025 • Created for educational purposes</p>
</div>
</div>
<script>
// Get the modal
var modal = document.getElementById("imageModal");
// Get the image and insert it inside the modal
var zoomableImages = document.getElementsByClassName("zoomable");
var modalImg = document.getElementById("zoomedImage");
// Add click event to all zoomable images
for (var i = 0; i < zoomableImages.length; i++) {
zoomableImages[i].onclick = function() {
modal.style.display = "flex";
modalImg.src = this.src;
}
}
// Get the <span> element that closes the modal
var span = document.getElementsByClassName("close")[0];
// When the user clicks on <span> (x), close the modal
span.onclick = function() {
modal.style.display = "none";
}
// Close the modal when clicking outside the image
modal.onclick = function(event) {
if (event.target === modal) {
modal.style.display = "none";
}
}
</script>
</body>
</html> |