Spaces:
Sleeping
Sleeping
Delete constants.py
Browse files- constants.py +0 -322
constants.py
DELETED
|
@@ -1,322 +0,0 @@
|
|
| 1 |
-
from collections import OrderedDict
|
| 2 |
-
|
| 3 |
-
BANNER = '''
|
| 4 |
-
<div style="
|
| 5 |
-
display: flex;
|
| 6 |
-
justify-content: center;
|
| 7 |
-
align-items: center;
|
| 8 |
-
height: 120px;
|
| 9 |
-
background: #ffffff;
|
| 10 |
-
border-bottom: 2px solid #ced4da;
|
| 11 |
-
">
|
| 12 |
-
<h1 style="
|
| 13 |
-
font-size: 5em;
|
| 14 |
-
font-family: 'Poppins', 'Segoe UI', sans-serif;
|
| 15 |
-
background: linear-gradient(90deg, #007bff, #00c6ff, #0096c7, #0077b6);
|
| 16 |
-
-webkit-background-clip: text;
|
| 17 |
-
color: transparent;
|
| 18 |
-
animation: gradientFlow 4s ease-in-out infinite;
|
| 19 |
-
background-size: 300%;
|
| 20 |
-
letter-spacing: 2px;
|
| 21 |
-
">
|
| 22 |
-
HardcoreLogic
|
| 23 |
-
</h1>
|
| 24 |
-
</div>
|
| 25 |
-
|
| 26 |
-
<style>
|
| 27 |
-
@keyframes gradientFlow {
|
| 28 |
-
0% { background-position: 0% 50%; }
|
| 29 |
-
50% { background-position: 100% 50%; }
|
| 30 |
-
100% { background-position: 0% 50%; }
|
| 31 |
-
}
|
| 32 |
-
</style>
|
| 33 |
-
'''
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
CITATION_TEXT = """
|
| 37 |
-
|
| 38 |
-
"""
|
| 39 |
-
'''
|
| 40 |
-
@article{hardcorelogic2025,
|
| 41 |
-
title={HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games},
|
| 42 |
-
author={},
|
| 43 |
-
year={2025},
|
| 44 |
-
url={https://arxiv.org/},
|
| 45 |
-
}'''
|
| 46 |
-
|
| 47 |
-
column_names = OrderedDict({
|
| 48 |
-
"model": "Model",
|
| 49 |
-
"open-source": "Open Source",
|
| 50 |
-
"total accuracy": "Total Acc",
|
| 51 |
-
"unsolvable puzzle": "Unsolvable Puzzle ACC",
|
| 52 |
-
})
|
| 53 |
-
|
| 54 |
-
column_names_puzzle = OrderedDict({
|
| 55 |
-
"model": "Model",
|
| 56 |
-
"total accuracy": "Total Acc",
|
| 57 |
-
"Zebra": "Zebra",
|
| 58 |
-
"Binario": "Binario",
|
| 59 |
-
"Crypto": "Crypto",
|
| 60 |
-
"Hanoi": "Hanoi",
|
| 61 |
-
"Hitpri": "Hitpri",
|
| 62 |
-
"Kakurasu": "Kakurasu",
|
| 63 |
-
"Minesweeper": "Minesweeper",
|
| 64 |
-
"Navigation": "Navigation",
|
| 65 |
-
"Skyscraper": "Skyscraper",
|
| 66 |
-
"Sudoku": "Sudoku",
|
| 67 |
-
})
|
| 68 |
-
|
| 69 |
-
LEADERBOARD_REMARKS = """**WB Reward**: for each comparison (A vs B), a reward for A is **+/-1** if A is **much better/worse** than B, and **+/-0.5** if A is **slightly better/worse** than B; when there is a **Tie**, the reward is **0**.
|
| 70 |
-
"""
|
| 71 |
-
|
| 72 |
-
LEADERBOARD_REMARKS_MAIN = """
|
| 73 |
-
"""
|
| 74 |
-
|
| 75 |
-
RANKING_COLUMN = "total accuracy"
|
| 76 |
-
|
| 77 |
-
ORDERED_COLUMN_NAMES = [
|
| 78 |
-
"model",
|
| 79 |
-
"mode",
|
| 80 |
-
"open-source",
|
| 81 |
-
"total accuracy",
|
| 82 |
-
"increased complexity",
|
| 83 |
-
"uncommon elements",
|
| 84 |
-
"unsolvable puzzle",
|
| 85 |
-
"temperature",
|
| 86 |
-
"n_sampling",
|
| 87 |
-
"n"
|
| 88 |
-
]
|
| 89 |
-
|
| 90 |
-
ORDERED_COLUMN_NAMES_PUZZLE = [
|
| 91 |
-
"model",
|
| 92 |
-
"mode",
|
| 93 |
-
"open-source",
|
| 94 |
-
"total accuracy",
|
| 95 |
-
"Zebra",
|
| 96 |
-
"Binario",
|
| 97 |
-
"Crypto",
|
| 98 |
-
"Hanoi",
|
| 99 |
-
"Hitpri",
|
| 100 |
-
"Kakurasu",
|
| 101 |
-
"Minesweeper",
|
| 102 |
-
"Navigation",
|
| 103 |
-
"Skyscraper",
|
| 104 |
-
"Sudoku",
|
| 105 |
-
"temperature",
|
| 106 |
-
"n_sampling",
|
| 107 |
-
"n"
|
| 108 |
-
]
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
js_light = """
|
| 112 |
-
function refresh() {
|
| 113 |
-
const url = new URL(window.location);
|
| 114 |
-
|
| 115 |
-
if (url.searchParams.get('__theme') !== 'light') {
|
| 116 |
-
url.searchParams.set('__theme', 'light');
|
| 117 |
-
window.location.href = url.href;
|
| 118 |
-
}
|
| 119 |
-
|
| 120 |
-
// Find the fieldset with the given id
|
| 121 |
-
const fieldset = document.getElementById("rank-column-radio");
|
| 122 |
-
|
| 123 |
-
// Create a new span element with the text "Decoding Mode:"
|
| 124 |
-
const rankBySpan = document.createElement("span");
|
| 125 |
-
rankBySpan.textContent = "Decoding Mode: ";
|
| 126 |
-
rankBySpan.style.fontWeight = "bold"; // Optional: make the text bold
|
| 127 |
-
rankBySpan.style.fontSize = "19px"; // Larger font size
|
| 128 |
-
rankBySpan.style.paddingRight = "18px"; // Add padding on the right
|
| 129 |
-
|
| 130 |
-
// Wrap the span and the labels in a flex container
|
| 131 |
-
const flexContainer = document.createElement("div");
|
| 132 |
-
flexContainer.style.display = "flex";
|
| 133 |
-
flexContainer.style.alignItems = "center";
|
| 134 |
-
|
| 135 |
-
// Insert the rankBySpan at the beginning of the flex container
|
| 136 |
-
flexContainer.appendChild(rankBySpan);
|
| 137 |
-
|
| 138 |
-
// Move all existing labels into the flex container
|
| 139 |
-
while (fieldset.firstChild) {
|
| 140 |
-
flexContainer.appendChild(fieldset.firstChild);
|
| 141 |
-
}
|
| 142 |
-
|
| 143 |
-
// Append the flex container back to the fieldset
|
| 144 |
-
fieldset.appendChild(flexContainer);
|
| 145 |
-
}
|
| 146 |
-
"""
|
| 147 |
-
|
| 148 |
-
js_code = """
|
| 149 |
-
function scroll_top() {
|
| 150 |
-
console.log("Hello from Gradio!");
|
| 151 |
-
const bubbles = document.querySelectorAll('.bubble-wrap');
|
| 152 |
-
bubbles.forEach((bubble, index) => {
|
| 153 |
-
setTimeout(() => {
|
| 154 |
-
bubble.scrollTop = 0;
|
| 155 |
-
}, index * 100); // Delay of 100ms between each iteration
|
| 156 |
-
});
|
| 157 |
-
|
| 158 |
-
}
|
| 159 |
-
"""
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
TASK_TYPE_STR = "**Tasks**: Info seeking (**InfoSek**), Creative Writing (**CrtWrt**), Coding&Debugging (**Code**), Reasoning (**Reason**), Editing (**Edit**), **Math**, Planning (**Plan**), Brainstorming (**Brnstrm**), Role playing (**RolPly**), Advice seeking (**AdvSek**), Data Analysis (**DataAna**)"
|
| 163 |
-
|
| 164 |
-
css = """
|
| 165 |
-
/* ========== 🌟 Global Typography ========== */
|
| 166 |
-
code {
|
| 167 |
-
font-size: large;
|
| 168 |
-
}
|
| 169 |
-
footer {visibility: hidden}
|
| 170 |
-
|
| 171 |
-
.markdown-text{font-size: 14pt}
|
| 172 |
-
.markdown-text-small{font-size: 13pt}
|
| 173 |
-
.markdown-text-tiny{font-size: 12pt}
|
| 174 |
-
|
| 175 |
-
/* ========== 🎓 Fudan Blue Theme Colors ========== */
|
| 176 |
-
:root {
|
| 177 |
-
--fudan-blue: #002D72;
|
| 178 |
-
--fudan-blue-light: #E6EEF8;
|
| 179 |
-
--fudan-gray: #f7f7f7;
|
| 180 |
-
--fudan-border: #c8d6e5;
|
| 181 |
-
--fudan-highlight: #1E56A0;
|
| 182 |
-
}
|
| 183 |
-
|
| 184 |
-
/* ========== 🏅 Leaderboard Table ========== */
|
| 185 |
-
#leaderboard-table th,
|
| 186 |
-
#leaderboard-puzzle-table th {
|
| 187 |
-
background-color: var(--fudan-blue);
|
| 188 |
-
color: white;
|
| 189 |
-
text-align: center;
|
| 190 |
-
padding: 10px;
|
| 191 |
-
font-size: 15px;
|
| 192 |
-
border-bottom: 2px solid var(--fudan-border);
|
| 193 |
-
}
|
| 194 |
-
|
| 195 |
-
#leaderboard-table td,
|
| 196 |
-
#leaderboard-puzzle-table td {
|
| 197 |
-
text-align: center;
|
| 198 |
-
font-size: 14px;
|
| 199 |
-
padding: 8px;
|
| 200 |
-
background-color: white;
|
| 201 |
-
border-bottom: 1px solid var(--fudan-border);
|
| 202 |
-
}
|
| 203 |
-
|
| 204 |
-
/* Hover 行高亮 */
|
| 205 |
-
#leaderboard-table tr:hover,
|
| 206 |
-
#leaderboard-puzzle-table tr:hover {
|
| 207 |
-
background-color: var(--fudan-blue-light);
|
| 208 |
-
transition: background-color 0.2s ease-in-out;
|
| 209 |
-
}
|
| 210 |
-
|
| 211 |
-
/* 表格整体外观 */
|
| 212 |
-
#leaderboard-table,
|
| 213 |
-
#leaderboard-puzzle-table {
|
| 214 |
-
border-collapse: collapse;
|
| 215 |
-
border-radius: 10px;
|
| 216 |
-
overflow: hidden;
|
| 217 |
-
box-shadow: 0 0 10px rgba(0, 45, 114, 0.15);
|
| 218 |
-
}
|
| 219 |
-
|
| 220 |
-
/* ========== 📊 Tabs ========== */
|
| 221 |
-
.tab-buttons button[role="tab"] {
|
| 222 |
-
font-size: 15px;
|
| 223 |
-
font-weight: 600;
|
| 224 |
-
color: var(--fudan-blue);
|
| 225 |
-
border: 1px solid var(--fudan-border);
|
| 226 |
-
border-radius: 8px;
|
| 227 |
-
background-color: white;
|
| 228 |
-
padding: 8px 16px;
|
| 229 |
-
margin-right: 5px;
|
| 230 |
-
transition: all 0.2s ease-in-out;
|
| 231 |
-
}
|
| 232 |
-
|
| 233 |
-
.tab-buttons button[role="tab"]:hover {
|
| 234 |
-
background-color: var(--fudan-blue-light);
|
| 235 |
-
}
|
| 236 |
-
|
| 237 |
-
button.selected[role="tab"][aria-selected="true"] {
|
| 238 |
-
background-color: var(--fudan-blue);
|
| 239 |
-
color: white;
|
| 240 |
-
font-weight: bold;
|
| 241 |
-
font-size: 16px;
|
| 242 |
-
}
|
| 243 |
-
|
| 244 |
-
/* ========== 📦 Accordion & Buttons ========== */
|
| 245 |
-
.accordion-label button span{
|
| 246 |
-
font-size: 14pt;
|
| 247 |
-
font-weight: bold;
|
| 248 |
-
color: var(--fudan-blue);
|
| 249 |
-
}
|
| 250 |
-
|
| 251 |
-
.btn_boderline{
|
| 252 |
-
border: 1px solid var(--fudan-blue);
|
| 253 |
-
border-radius: 5px;
|
| 254 |
-
padding: 6px 12px;
|
| 255 |
-
margin: 5px;
|
| 256 |
-
font-size: 14pt;
|
| 257 |
-
font-weight: bold;
|
| 258 |
-
background-color: var(--fudan-blue-light);
|
| 259 |
-
color: var(--fudan-blue);
|
| 260 |
-
transition: background-color 0.3s;
|
| 261 |
-
}
|
| 262 |
-
.btn_boderline:hover{
|
| 263 |
-
background-color: var(--fudan-blue);
|
| 264 |
-
color: white;
|
| 265 |
-
}
|
| 266 |
-
|
| 267 |
-
/* ========== 🧩 Box & Card ========== */
|
| 268 |
-
.box_md{
|
| 269 |
-
border: 1px solid var(--fudan-border);
|
| 270 |
-
border-radius: 10px;
|
| 271 |
-
padding: 10px;
|
| 272 |
-
font-size: 12pt;
|
| 273 |
-
margin: 8px;
|
| 274 |
-
background-color: white;
|
| 275 |
-
box-shadow: 0 0 6px rgba(0, 45, 114, 0.1);
|
| 276 |
-
}
|
| 277 |
-
|
| 278 |
-
/* ========== 💬 Markdown Text Enhancements ========== */
|
| 279 |
-
.markdown-text-details{
|
| 280 |
-
margin: 10px;
|
| 281 |
-
padding: 10px;
|
| 282 |
-
background-color: var(--fudan-gray);
|
| 283 |
-
border-left: 4px solid var(--fudan-blue);
|
| 284 |
-
border-radius: 6px;
|
| 285 |
-
}
|
| 286 |
-
|
| 287 |
-
/* ========== 📈 Plot & Visualization ========== */
|
| 288 |
-
.plotly-plot{
|
| 289 |
-
height: auto;
|
| 290 |
-
max-height: 600px;
|
| 291 |
-
min-height: 600px;
|
| 292 |
-
border: 1px solid var(--fudan-border);
|
| 293 |
-
border-radius: 10px;
|
| 294 |
-
}
|
| 295 |
-
|
| 296 |
-
/* ========== 🧷 Misc Components ========== */
|
| 297 |
-
.sample_button{
|
| 298 |
-
border: 2px solid var(--fudan-blue);
|
| 299 |
-
border-radius: 10px;
|
| 300 |
-
padding: 10px;
|
| 301 |
-
font-size: 17pt;
|
| 302 |
-
font-weight: bold;
|
| 303 |
-
margin: 5px;
|
| 304 |
-
background-color: var(--fudan-blue-light);
|
| 305 |
-
color: var(--fudan-blue);
|
| 306 |
-
transition: all 0.3s ease-in-out;
|
| 307 |
-
}
|
| 308 |
-
.sample_button:hover {
|
| 309 |
-
background-color: var(--fudan-blue);
|
| 310 |
-
color: white;
|
| 311 |
-
}
|
| 312 |
-
|
| 313 |
-
/* Scrollable Table Containers */
|
| 314 |
-
#leaderboard-table,
|
| 315 |
-
#leaderboard-puzzle-table {
|
| 316 |
-
display: block;
|
| 317 |
-
max-height: 800px;
|
| 318 |
-
overflow-y: auto;
|
| 319 |
-
}
|
| 320 |
-
|
| 321 |
-
|
| 322 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|