UI improvements for light mode
Browse files
app.py
CHANGED
|
@@ -1222,7 +1222,7 @@ class ProgressTracker:
|
|
| 1222 |
# Add model information if available and we're not in idle or error state
|
| 1223 |
model_info = ''
|
| 1224 |
if self.stage not in ["idle", "error", "starting"] and (self.generator_model or self.judge_model):
|
| 1225 |
-
model_info = f'<div class="model-info"
|
| 1226 |
if self.generator_model:
|
| 1227 |
model_info += f'<div><span style="font-weight: bold;">Generator:</span> {self.generator_model}</div>'
|
| 1228 |
if self.judge_model:
|
|
@@ -1293,21 +1293,24 @@ def create_interface():
|
|
| 1293 |
|
| 1294 |
# CSS for styling
|
| 1295 |
css = """
|
|
|
|
| 1296 |
.container {
|
| 1297 |
max-width: 1000px;
|
| 1298 |
margin: 0 auto;
|
| 1299 |
}
|
|
|
|
|
|
|
| 1300 |
.title {
|
| 1301 |
text-align: center;
|
| 1302 |
margin-bottom: 0.5em;
|
| 1303 |
-
color: #0d47a1;
|
| 1304 |
font-weight: 600;
|
|
|
|
| 1305 |
}
|
| 1306 |
.subtitle {
|
| 1307 |
text-align: center;
|
| 1308 |
margin-bottom: 1.5em;
|
| 1309 |
-
color: #37474f;
|
| 1310 |
font-size: 1.2em;
|
|
|
|
| 1311 |
}
|
| 1312 |
.section-title {
|
| 1313 |
margin-top: 1em;
|
|
@@ -1318,63 +1321,63 @@ def create_interface():
|
|
| 1318 |
.info-box {
|
| 1319 |
padding: 1.2em;
|
| 1320 |
border-radius: 8px;
|
| 1321 |
-
background-color: var(--background-fill-secondary);
|
| 1322 |
margin-bottom: 1em;
|
| 1323 |
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
| 1324 |
-
color: var(--body-text-color);
|
| 1325 |
line-height: 1.5;
|
| 1326 |
-
border
|
| 1327 |
-
border:
|
|
|
|
|
|
|
| 1328 |
}
|
| 1329 |
.info-box p strong {
|
| 1330 |
-
color:
|
| 1331 |
font-weight: 600;
|
| 1332 |
}
|
| 1333 |
.hallucination-positive {
|
| 1334 |
padding: 1.2em;
|
| 1335 |
border-radius: 8px;
|
| 1336 |
-
background-color: #
|
| 1337 |
-
border-left: 5px solid #
|
| 1338 |
margin-bottom: 1em;
|
| 1339 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1340 |
-
color: #
|
| 1341 |
}
|
| 1342 |
.hallucination-positive h3 {
|
| 1343 |
-
color: #
|
| 1344 |
margin-top: 0;
|
| 1345 |
margin-bottom: 0.5em;
|
| 1346 |
}
|
| 1347 |
.hallucination-positive p {
|
| 1348 |
-
color: #
|
| 1349 |
line-height: 1.5;
|
| 1350 |
}
|
| 1351 |
.hallucination-negative {
|
| 1352 |
padding: 1.2em;
|
| 1353 |
border-radius: 8px;
|
| 1354 |
-
background-color: #
|
| 1355 |
-
border-left: 5px solid #
|
| 1356 |
margin-bottom: 1em;
|
| 1357 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1358 |
-
color: #
|
| 1359 |
}
|
| 1360 |
.hallucination-negative h3 {
|
| 1361 |
-
color: #
|
| 1362 |
margin-top: 0;
|
| 1363 |
margin-bottom: 0.5em;
|
| 1364 |
}
|
| 1365 |
.hallucination-negative p {
|
| 1366 |
-
color: #
|
| 1367 |
line-height: 1.5;
|
| 1368 |
}
|
| 1369 |
.response-box {
|
| 1370 |
padding: 1.2em;
|
| 1371 |
border-radius: 8px;
|
| 1372 |
-
background-color: #
|
| 1373 |
margin-bottom: 0.8em;
|
| 1374 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1375 |
-
color: #
|
| 1376 |
line-height: 1.5;
|
| 1377 |
-
border-left: 3px solid #
|
| 1378 |
}
|
| 1379 |
.example-queries {
|
| 1380 |
display: flex;
|
|
@@ -1383,42 +1386,46 @@ def create_interface():
|
|
| 1383 |
margin-bottom: 15px;
|
| 1384 |
}
|
| 1385 |
.example-query {
|
| 1386 |
-
background-color: #
|
| 1387 |
padding: 8px 15px;
|
| 1388 |
border-radius: 18px;
|
| 1389 |
font-size: 0.9em;
|
| 1390 |
cursor: pointer;
|
| 1391 |
transition: all 0.2s;
|
| 1392 |
-
border: 1px solid #
|
| 1393 |
-
color: #
|
| 1394 |
}
|
| 1395 |
.example-query:hover {
|
| 1396 |
-
background-color: #
|
| 1397 |
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
| 1398 |
}
|
| 1399 |
.stats-section {
|
| 1400 |
display: flex;
|
| 1401 |
justify-content: space-between;
|
| 1402 |
-
background-color: #
|
| 1403 |
padding: 15px;
|
| 1404 |
-
border-radius:
|
| 1405 |
margin-bottom: 20px;
|
|
|
|
|
|
|
|
|
|
| 1406 |
}
|
| 1407 |
.stat-item {
|
| 1408 |
text-align: center;
|
| 1409 |
padding: 10px;
|
| 1410 |
}
|
| 1411 |
.stat-value {
|
| 1412 |
-
font-size:
|
| 1413 |
font-weight: bold;
|
| 1414 |
-
color: #
|
| 1415 |
}
|
| 1416 |
.stat-label {
|
| 1417 |
font-size: 0.9em;
|
| 1418 |
-
|
|
|
|
| 1419 |
}
|
| 1420 |
.feedback-section {
|
| 1421 |
-
border-top: 1px solid #
|
| 1422 |
padding-top: 15px;
|
| 1423 |
margin-top: 20px;
|
| 1424 |
}
|
|
@@ -1426,20 +1433,20 @@ def create_interface():
|
|
| 1426 |
text-align: center;
|
| 1427 |
padding: 20px;
|
| 1428 |
margin-top: 30px;
|
| 1429 |
-
color: #
|
| 1430 |
font-size: 0.9em;
|
| 1431 |
}
|
| 1432 |
.processing-status {
|
| 1433 |
padding: 12px;
|
| 1434 |
-
background-color: #
|
| 1435 |
-
border-left: 4px solid #
|
| 1436 |
margin-bottom: 15px;
|
| 1437 |
font-weight: 500;
|
| 1438 |
-
color: #
|
| 1439 |
}
|
| 1440 |
.debug-panel {
|
| 1441 |
-
background-color: #
|
| 1442 |
-
border: 1px solid #
|
| 1443 |
border-radius: 4px;
|
| 1444 |
padding: 10px;
|
| 1445 |
margin-top: 15px;
|
|
@@ -1448,13 +1455,15 @@ def create_interface():
|
|
| 1448 |
white-space: pre-wrap;
|
| 1449 |
max-height: 200px;
|
| 1450 |
overflow-y: auto;
|
|
|
|
| 1451 |
}
|
| 1452 |
.progress-container {
|
| 1453 |
padding: 15px;
|
| 1454 |
-
background-color: #
|
| 1455 |
border-radius: 8px;
|
| 1456 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1457 |
margin-bottom: 15px;
|
|
|
|
| 1458 |
}
|
| 1459 |
.progress-status {
|
| 1460 |
font-weight: 500;
|
|
@@ -1463,7 +1472,7 @@ def create_interface():
|
|
| 1463 |
font-size: 0.95em;
|
| 1464 |
}
|
| 1465 |
.progress-bar-container {
|
| 1466 |
-
background-color: #
|
| 1467 |
height: 10px;
|
| 1468 |
border-radius: 5px;
|
| 1469 |
overflow: hidden;
|
|
@@ -1473,16 +1482,595 @@ def create_interface():
|
|
| 1473 |
.progress-bar {
|
| 1474 |
height: 100%;
|
| 1475 |
transition: width 0.5s ease;
|
| 1476 |
-
background-image: linear-gradient(to right, #
|
| 1477 |
}
|
| 1478 |
.query-display {
|
| 1479 |
font-style: italic;
|
| 1480 |
-
color: #
|
| 1481 |
margin-bottom: 10px;
|
| 1482 |
-
background-color: #
|
| 1483 |
padding: 8px;
|
| 1484 |
border-radius: 4px;
|
| 1485 |
-
border-left: 3px solid #
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1486 |
}
|
| 1487 |
"""
|
| 1488 |
|
|
@@ -1749,14 +2337,14 @@ def create_interface():
|
|
| 1749 |
<div class="container">
|
| 1750 |
<h2 class="title">Hallucination Detection Results</h2>
|
| 1751 |
|
| 1752 |
-
<div class="model-info-bar"
|
| 1753 |
-
<div
|
| 1754 |
-
<div
|
| 1755 |
-
<div
|
| 1756 |
</div>
|
| 1757 |
-
<div
|
| 1758 |
-
<div
|
| 1759 |
-
<div
|
| 1760 |
</div>
|
| 1761 |
</div>
|
| 1762 |
|
|
@@ -1789,7 +2377,7 @@ def create_interface():
|
|
| 1789 |
{original_query}
|
| 1790 |
</div>
|
| 1791 |
|
| 1792 |
-
<div class="section-title">Original Response <span
|
| 1793 |
<div class="response-box">
|
| 1794 |
{original_response_safe}
|
| 1795 |
</div>
|
|
@@ -1804,14 +2392,14 @@ def create_interface():
|
|
| 1804 |
{q}
|
| 1805 |
</div>
|
| 1806 |
|
| 1807 |
-
<div class="section-title">Response {i} <span
|
| 1808 |
<div class="response-box">
|
| 1809 |
{r}
|
| 1810 |
</div>
|
| 1811 |
"""
|
| 1812 |
|
| 1813 |
html_output += f"""
|
| 1814 |
-
<div class="section-title">Detailed Analysis <span
|
| 1815 |
<div class="info-box">
|
| 1816 |
<p><strong>Reasoning:</strong></p>
|
| 1817 |
<p>{reasoning_safe}</p>
|
|
@@ -1820,7 +2408,7 @@ def create_interface():
|
|
| 1820 |
<p>{conflicting_facts_text_safe}</p>
|
| 1821 |
</div>
|
| 1822 |
|
| 1823 |
-
<div
|
| 1824 |
Models randomly selected for this analysis: <strong>{generator_model}</strong> (Generator) and <strong>{judge_model}</strong> (Judge)
|
| 1825 |
</div>
|
| 1826 |
</div>
|
|
@@ -1945,9 +2533,9 @@ def create_interface():
|
|
| 1945 |
gr.HTML(
|
| 1946 |
"""
|
| 1947 |
<div style="text-align: center; margin-bottom: 1.5rem">
|
| 1948 |
-
<h1
|
| 1949 |
-
<h3
|
| 1950 |
-
<p
|
| 1951 |
This tool detects hallucinations in AI responses by comparing answers to semantically equivalent questions and using a specialized judge model.
|
| 1952 |
</p>
|
| 1953 |
</div>
|
|
@@ -2075,9 +2663,8 @@ def create_interface():
|
|
| 2075 |
|
| 2076 |
if not pairs:
|
| 2077 |
return (
|
| 2078 |
-
"<div class=\"info-message\"
|
| 2079 |
-
"
|
| 2080 |
-
"<h3 style=\"margin-top: 0; color: #0277bd;\">No Data Available Yet</h3>"
|
| 2081 |
"<p>Try the detector with more queries to populate the leaderboard!</p>"
|
| 2082 |
"</div>"
|
| 2083 |
)
|
|
@@ -2106,9 +2693,9 @@ def create_interface():
|
|
| 2106 |
f"<td>{pair.get('judge', 'unknown')}</td>"
|
| 2107 |
f"<td>{round(pair.get('elo_score', 0))}</td>"
|
| 2108 |
f"<td>{pair.get('accuracy')}%</td>"
|
| 2109 |
-
f"<td
|
| 2110 |
-
f"<td
|
| 2111 |
-
f"<td
|
| 2112 |
f"<td>{pair.get('total_samples', 0)}</td>"
|
| 2113 |
f"</tr>"
|
| 2114 |
)
|
|
@@ -2135,13 +2722,13 @@ def create_interface():
|
|
| 2135 |
f"</tbody>"
|
| 2136 |
f"</table>"
|
| 2137 |
f"</div>"
|
| 2138 |
-
f"<div
|
| 2139 |
-
f"<p
|
| 2140 |
-
f"<ul
|
| 2141 |
-
f"<li><strong
|
| 2142 |
-
f"<li><strong
|
| 2143 |
-
f"<li><strong
|
| 2144 |
-
f"<li><strong
|
| 2145 |
f"</ul>"
|
| 2146 |
f"</div>"
|
| 2147 |
)
|
|
@@ -2150,9 +2737,8 @@ def create_interface():
|
|
| 2150 |
except Exception as e:
|
| 2151 |
logger.error("Error generating leaderboard HTML: %s", str(e), exc_info=True)
|
| 2152 |
return (
|
| 2153 |
-
f"<div class=\"error-message\"
|
| 2154 |
-
f"
|
| 2155 |
-
f"<h3 style=\"margin-top: 0; color: #c62828;\">Error Loading Leaderboard</h3>"
|
| 2156 |
f"<p>{str(e)}</p>"
|
| 2157 |
f"</div>"
|
| 2158 |
)
|
|
@@ -2192,24 +2778,6 @@ def create_interface():
|
|
| 2192 |
"<li>deepseek-reasoner</li><li>o4-mini</li><li>gemini-2.5-pro</li>" +
|
| 2193 |
"</ul></div></div></div></div></div>"
|
| 2194 |
)
|
| 2195 |
-
gr.HTML(
|
| 2196 |
-
"<style>" +
|
| 2197 |
-
".leaderboard-container {margin: 15px 0; overflow-x: auto;}" +
|
| 2198 |
-
".leaderboard-table {width: 100%; border-collapse: collapse; font-size: 0.95em; " +
|
| 2199 |
-
"box-shadow: 0 2px 10px rgba(0,0,0,0.2); border-radius: 8px; overflow: hidden;}" +
|
| 2200 |
-
".leaderboard-table thead {background-color: #0d47a1; color: white;}" +
|
| 2201 |
-
".leaderboard-table th, .leaderboard-table td {padding: 12px 15px; text-align: left; border-bottom: 1px solid #37474f; color: #eceff1;}" +
|
| 2202 |
-
".leaderboard-table tbody tr {transition: background-color 0.3s;}" +
|
| 2203 |
-
".leaderboard-table tbody tr:nth-child(even) {background-color: #37474f;}" +
|
| 2204 |
-
".leaderboard-table tbody tr:nth-child(odd) {background-color: #455a64;}" +
|
| 2205 |
-
".leaderboard-table tbody tr:hover {background-color: #263238;}" +
|
| 2206 |
-
".leaderboard-table tbody tr.top-rank-1 {background-color: #004d40; color: #e0f2f1; font-weight: bold;}" +
|
| 2207 |
-
".leaderboard-table tbody tr.top-rank-2 {background-color: #1b5e20; color: #e8f5e9; font-weight: 500;}" +
|
| 2208 |
-
".leaderboard-table tbody tr.top-rank-3 {background-color: #33691e; color: #f1f8e9; font-weight: 500;}" +
|
| 2209 |
-
".leaderboard-table td {position: relative;}" +
|
| 2210 |
-
".leaderboard-table td::after {content: ''; position: absolute; top: 0; left: 0; width: 100%; height: 100%; background: transparent; pointer-events: none;}" +
|
| 2211 |
-
"</style>"
|
| 2212 |
-
)
|
| 2213 |
|
| 2214 |
# Tab 3: Individual Models Leaderboard
|
| 2215 |
with gr.TabItem("Individual Models", elem_id="user-feedback-tab"):
|
|
@@ -2224,9 +2792,8 @@ def create_interface():
|
|
| 2224 |
|
| 2225 |
if not models:
|
| 2226 |
return (
|
| 2227 |
-
"<div class=\"info-message\"
|
| 2228 |
-
"
|
| 2229 |
-
"<h3 style=\"margin-top: 0; color: #0277bd;\">No Data Available Yet</h3>"
|
| 2230 |
"<p>Try the detector with more queries to populate the model scores!</p>"
|
| 2231 |
"</div>"
|
| 2232 |
)
|
|
@@ -2263,10 +2830,10 @@ def create_interface():
|
|
| 2263 |
f"<td>{model.get('model_name', 'unknown')}</td>"
|
| 2264 |
f"<td>{round(model.get('elo_score', 0))}</td>"
|
| 2265 |
f"<td>{model.get('accuracy')}%</td>"
|
| 2266 |
-
f"<td
|
| 2267 |
-
f"<td
|
| 2268 |
f"<td>{model.get('total_samples', 0)}</td>"
|
| 2269 |
-
f"<td
|
| 2270 |
f"</tr>"
|
| 2271 |
)
|
| 2272 |
|
|
@@ -2297,9 +2864,8 @@ def create_interface():
|
|
| 2297 |
except Exception as e:
|
| 2298 |
logger.error("Error generating model leaderboard HTML: %s", str(e), exc_info=True)
|
| 2299 |
return (
|
| 2300 |
-
f"<div class=\"error-message\"
|
| 2301 |
-
f"
|
| 2302 |
-
f"<h3 style=\"margin-top: 0; color: #c62828;\">Error Loading Model Leaderboard</h3>"
|
| 2303 |
f"<p>{str(e)}</p>"
|
| 2304 |
f"</div>"
|
| 2305 |
)
|
|
@@ -2356,17 +2922,17 @@ def create_interface():
|
|
| 2356 |
accuracy_pct = f"{accuracy * 100:.1f}%"
|
| 2357 |
|
| 2358 |
stats_html = f"""
|
| 2359 |
-
<div class="stats-section"
|
| 2360 |
<div class="stat-item">
|
| 2361 |
-
<div class="stat-value"
|
| 2362 |
-
<div class="stat-label"
|
| 2363 |
</div>
|
| 2364 |
<div class="stat-item">
|
| 2365 |
-
<div class="stat-value"
|
| 2366 |
-
<div class="stat-label"
|
| 2367 |
</div>
|
| 2368 |
</div>
|
| 2369 |
-
<div style="text-align: center; margin-top: 10px; font-style: italic;
|
| 2370 |
Based on user feedback: {correct} correct out of {total} total predictions
|
| 2371 |
</div>
|
| 2372 |
"""
|
|
|
|
| 1222 |
# Add model information if available and we're not in idle or error state
|
| 1223 |
model_info = ''
|
| 1224 |
if self.stage not in ["idle", "error", "starting"] and (self.generator_model or self.judge_model):
|
| 1225 |
+
model_info = f'<div class="progress-model-info">'
|
| 1226 |
if self.generator_model:
|
| 1227 |
model_info += f'<div><span style="font-weight: bold;">Generator:</span> {self.generator_model}</div>'
|
| 1228 |
if self.judge_model:
|
|
|
|
| 1293 |
|
| 1294 |
# CSS for styling
|
| 1295 |
css = """
|
| 1296 |
+
/* Base styles */
|
| 1297 |
.container {
|
| 1298 |
max-width: 1000px;
|
| 1299 |
margin: 0 auto;
|
| 1300 |
}
|
| 1301 |
+
|
| 1302 |
+
/* Light theme default styles */
|
| 1303 |
.title {
|
| 1304 |
text-align: center;
|
| 1305 |
margin-bottom: 0.5em;
|
|
|
|
| 1306 |
font-weight: 600;
|
| 1307 |
+
color: #0d47a1;
|
| 1308 |
}
|
| 1309 |
.subtitle {
|
| 1310 |
text-align: center;
|
| 1311 |
margin-bottom: 1.5em;
|
|
|
|
| 1312 |
font-size: 1.2em;
|
| 1313 |
+
color: #37474f;
|
| 1314 |
}
|
| 1315 |
.section-title {
|
| 1316 |
margin-top: 1em;
|
|
|
|
| 1321 |
.info-box {
|
| 1322 |
padding: 1.2em;
|
| 1323 |
border-radius: 8px;
|
|
|
|
| 1324 |
margin-bottom: 1em;
|
| 1325 |
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
|
|
|
| 1326 |
line-height: 1.5;
|
| 1327 |
+
border: 1px solid #dee2e6;
|
| 1328 |
+
border-left: 3px solid #6c757d;
|
| 1329 |
+
background-color: #f8f9fa;
|
| 1330 |
+
color: #212529;
|
| 1331 |
}
|
| 1332 |
.info-box p strong {
|
| 1333 |
+
color: #495057;
|
| 1334 |
font-weight: 600;
|
| 1335 |
}
|
| 1336 |
.hallucination-positive {
|
| 1337 |
padding: 1.2em;
|
| 1338 |
border-radius: 8px;
|
| 1339 |
+
background-color: #ffeaea;
|
| 1340 |
+
border-left: 5px solid #e53e3e;
|
| 1341 |
margin-bottom: 1em;
|
| 1342 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1343 |
+
color: #742a2a;
|
| 1344 |
}
|
| 1345 |
.hallucination-positive h3 {
|
| 1346 |
+
color: #e53e3e;
|
| 1347 |
margin-top: 0;
|
| 1348 |
margin-bottom: 0.5em;
|
| 1349 |
}
|
| 1350 |
.hallucination-positive p {
|
| 1351 |
+
color: #742a2a;
|
| 1352 |
line-height: 1.5;
|
| 1353 |
}
|
| 1354 |
.hallucination-negative {
|
| 1355 |
padding: 1.2em;
|
| 1356 |
border-radius: 8px;
|
| 1357 |
+
background-color: #f0fff4;
|
| 1358 |
+
border-left: 5px solid #38a169;
|
| 1359 |
margin-bottom: 1em;
|
| 1360 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1361 |
+
color: #22543d;
|
| 1362 |
}
|
| 1363 |
.hallucination-negative h3 {
|
| 1364 |
+
color: #38a169;
|
| 1365 |
margin-top: 0;
|
| 1366 |
margin-bottom: 0.5em;
|
| 1367 |
}
|
| 1368 |
.hallucination-negative p {
|
| 1369 |
+
color: #22543d;
|
| 1370 |
line-height: 1.5;
|
| 1371 |
}
|
| 1372 |
.response-box {
|
| 1373 |
padding: 1.2em;
|
| 1374 |
border-radius: 8px;
|
| 1375 |
+
background-color: #f7fafc;
|
| 1376 |
margin-bottom: 0.8em;
|
| 1377 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1378 |
+
color: #2d3748;
|
| 1379 |
line-height: 1.5;
|
| 1380 |
+
border-left: 3px solid #a0aec0;
|
| 1381 |
}
|
| 1382 |
.example-queries {
|
| 1383 |
display: flex;
|
|
|
|
| 1386 |
margin-bottom: 15px;
|
| 1387 |
}
|
| 1388 |
.example-query {
|
| 1389 |
+
background-color: #ebf8ff;
|
| 1390 |
padding: 8px 15px;
|
| 1391 |
border-radius: 18px;
|
| 1392 |
font-size: 0.9em;
|
| 1393 |
cursor: pointer;
|
| 1394 |
transition: all 0.2s;
|
| 1395 |
+
border: 1px solid #bee3f8;
|
| 1396 |
+
color: #2c5282;
|
| 1397 |
}
|
| 1398 |
.example-query:hover {
|
| 1399 |
+
background-color: #bee3f8;
|
| 1400 |
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
| 1401 |
}
|
| 1402 |
.stats-section {
|
| 1403 |
display: flex;
|
| 1404 |
justify-content: space-between;
|
| 1405 |
+
background-color: #ebf8ff;
|
| 1406 |
padding: 15px;
|
| 1407 |
+
border-radius: 10px;
|
| 1408 |
margin-bottom: 20px;
|
| 1409 |
+
margin-top: 5px;
|
| 1410 |
+
box-shadow: 0 2px 10px rgba(0,0,0,0.1);
|
| 1411 |
+
border: 1px solid #bee3f8;
|
| 1412 |
}
|
| 1413 |
.stat-item {
|
| 1414 |
text-align: center;
|
| 1415 |
padding: 10px;
|
| 1416 |
}
|
| 1417 |
.stat-value {
|
| 1418 |
+
font-size: 2em;
|
| 1419 |
font-weight: bold;
|
| 1420 |
+
color: #2c5282;
|
| 1421 |
}
|
| 1422 |
.stat-label {
|
| 1423 |
font-size: 0.9em;
|
| 1424 |
+
font-weight: bold;
|
| 1425 |
+
color: #3182ce;
|
| 1426 |
}
|
| 1427 |
.feedback-section {
|
| 1428 |
+
border-top: 1px solid #e2e8f0;
|
| 1429 |
padding-top: 15px;
|
| 1430 |
margin-top: 20px;
|
| 1431 |
}
|
|
|
|
| 1433 |
text-align: center;
|
| 1434 |
padding: 20px;
|
| 1435 |
margin-top: 30px;
|
| 1436 |
+
color: #718096;
|
| 1437 |
font-size: 0.9em;
|
| 1438 |
}
|
| 1439 |
.processing-status {
|
| 1440 |
padding: 12px;
|
| 1441 |
+
background-color: #ebf8ff;
|
| 1442 |
+
border-left: 4px solid #3182ce;
|
| 1443 |
margin-bottom: 15px;
|
| 1444 |
font-weight: 500;
|
| 1445 |
+
color: #2c5282;
|
| 1446 |
}
|
| 1447 |
.debug-panel {
|
| 1448 |
+
background-color: #f7fafc;
|
| 1449 |
+
border: 1px solid #e2e8f0;
|
| 1450 |
border-radius: 4px;
|
| 1451 |
padding: 10px;
|
| 1452 |
margin-top: 15px;
|
|
|
|
| 1455 |
white-space: pre-wrap;
|
| 1456 |
max-height: 200px;
|
| 1457 |
overflow-y: auto;
|
| 1458 |
+
color: #4a5568;
|
| 1459 |
}
|
| 1460 |
.progress-container {
|
| 1461 |
padding: 15px;
|
| 1462 |
+
background-color: #ffffff;
|
| 1463 |
border-radius: 8px;
|
| 1464 |
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 1465 |
margin-bottom: 15px;
|
| 1466 |
+
border: 1px solid #e2e8f0;
|
| 1467 |
}
|
| 1468 |
.progress-status {
|
| 1469 |
font-weight: 500;
|
|
|
|
| 1472 |
font-size: 0.95em;
|
| 1473 |
}
|
| 1474 |
.progress-bar-container {
|
| 1475 |
+
background-color: #edf2f7;
|
| 1476 |
height: 10px;
|
| 1477 |
border-radius: 5px;
|
| 1478 |
overflow: hidden;
|
|
|
|
| 1482 |
.progress-bar {
|
| 1483 |
height: 100%;
|
| 1484 |
transition: width 0.5s ease;
|
| 1485 |
+
background-image: linear-gradient(to right, #3182ce, #2b6cb0);
|
| 1486 |
}
|
| 1487 |
.query-display {
|
| 1488 |
font-style: italic;
|
| 1489 |
+
color: #718096;
|
| 1490 |
margin-bottom: 10px;
|
| 1491 |
+
background-color: #f7fafc;
|
| 1492 |
padding: 8px;
|
| 1493 |
border-radius: 4px;
|
| 1494 |
+
border-left: 3px solid #3182ce;
|
| 1495 |
+
}
|
| 1496 |
+
|
| 1497 |
+
/* Dark theme styles */
|
| 1498 |
+
@media (prefers-color-scheme: dark) {
|
| 1499 |
+
.title {
|
| 1500 |
+
color: #63b3ed;
|
| 1501 |
+
}
|
| 1502 |
+
.subtitle {
|
| 1503 |
+
color: #a0aec0;
|
| 1504 |
+
}
|
| 1505 |
+
.section-title {
|
| 1506 |
+
color: #90cdf4;
|
| 1507 |
+
}
|
| 1508 |
+
.info-box {
|
| 1509 |
+
background-color: #2d3748;
|
| 1510 |
+
color: #e2e8f0;
|
| 1511 |
+
border-color: #4a5568;
|
| 1512 |
+
border-left-color: #718096;
|
| 1513 |
+
}
|
| 1514 |
+
.info-box p strong {
|
| 1515 |
+
color: #f7fafc;
|
| 1516 |
+
}
|
| 1517 |
+
.hallucination-positive {
|
| 1518 |
+
background-color: #553c39;
|
| 1519 |
+
color: #fed7d7;
|
| 1520 |
+
border-left-color: #fc8181;
|
| 1521 |
+
}
|
| 1522 |
+
.hallucination-positive h3 {
|
| 1523 |
+
color: #fc8181;
|
| 1524 |
+
}
|
| 1525 |
+
.hallucination-positive p {
|
| 1526 |
+
color: #fed7d7;
|
| 1527 |
+
}
|
| 1528 |
+
.hallucination-negative {
|
| 1529 |
+
background-color: #22543d;
|
| 1530 |
+
color: #c6f6d5;
|
| 1531 |
+
border-left-color: #68d391;
|
| 1532 |
+
}
|
| 1533 |
+
.hallucination-negative h3 {
|
| 1534 |
+
color: #68d391;
|
| 1535 |
+
}
|
| 1536 |
+
.hallucination-negative p {
|
| 1537 |
+
color: #c6f6d5;
|
| 1538 |
+
}
|
| 1539 |
+
.response-box {
|
| 1540 |
+
background-color: #1a202c;
|
| 1541 |
+
color: #e2e8f0;
|
| 1542 |
+
border-left-color: #4a5568;
|
| 1543 |
+
}
|
| 1544 |
+
.example-query {
|
| 1545 |
+
background-color: #2a4365;
|
| 1546 |
+
border-color: #2c5282;
|
| 1547 |
+
color: #bee3f8;
|
| 1548 |
+
}
|
| 1549 |
+
.example-query:hover {
|
| 1550 |
+
background-color: #3182ce;
|
| 1551 |
+
}
|
| 1552 |
+
.stats-section {
|
| 1553 |
+
background-color: #2a4365;
|
| 1554 |
+
border-color: #2c5282;
|
| 1555 |
+
}
|
| 1556 |
+
.stat-value {
|
| 1557 |
+
color: #bee3f8;
|
| 1558 |
+
}
|
| 1559 |
+
.stat-label {
|
| 1560 |
+
color: #90cdf4;
|
| 1561 |
+
}
|
| 1562 |
+
.feedback-section {
|
| 1563 |
+
border-top-color: #4a5568;
|
| 1564 |
+
}
|
| 1565 |
+
.footer {
|
| 1566 |
+
color: #a0aec0;
|
| 1567 |
+
}
|
| 1568 |
+
.processing-status {
|
| 1569 |
+
background-color: #2a4365;
|
| 1570 |
+
border-left-color: #90cdf4;
|
| 1571 |
+
color: #bee3f8;
|
| 1572 |
+
}
|
| 1573 |
+
.debug-panel {
|
| 1574 |
+
background-color: #1a202c;
|
| 1575 |
+
border-color: #4a5568;
|
| 1576 |
+
color: #e2e8f0;
|
| 1577 |
+
}
|
| 1578 |
+
.progress-container {
|
| 1579 |
+
background-color: #2d3748;
|
| 1580 |
+
border-color: #4a5568;
|
| 1581 |
+
}
|
| 1582 |
+
.progress-bar-container {
|
| 1583 |
+
background-color: #4a5568;
|
| 1584 |
+
}
|
| 1585 |
+
.progress-bar {
|
| 1586 |
+
background-image: linear-gradient(to right, #90cdf4, #63b3ed);
|
| 1587 |
+
}
|
| 1588 |
+
.query-display {
|
| 1589 |
+
color: #a0aec0;
|
| 1590 |
+
background-color: #1a202c;
|
| 1591 |
+
border-left-color: #90cdf4;
|
| 1592 |
+
}
|
| 1593 |
+
}
|
| 1594 |
+
|
| 1595 |
+
/* Gradio theme detection fallbacks */
|
| 1596 |
+
html[data-theme="dark"] .title,
|
| 1597 |
+
.dark .title {
|
| 1598 |
+
color: #63b3ed !important;
|
| 1599 |
+
}
|
| 1600 |
+
html[data-theme="dark"] .subtitle,
|
| 1601 |
+
.dark .subtitle {
|
| 1602 |
+
color: #a0aec0 !important;
|
| 1603 |
+
}
|
| 1604 |
+
html[data-theme="dark"] .section-title,
|
| 1605 |
+
.dark .section-title {
|
| 1606 |
+
color: #90cdf4 !important;
|
| 1607 |
+
}
|
| 1608 |
+
html[data-theme="dark"] .info-box,
|
| 1609 |
+
.dark .info-box {
|
| 1610 |
+
background-color: #2d3748 !important;
|
| 1611 |
+
color: #e2e8f0 !important;
|
| 1612 |
+
border-color: #4a5568 !important;
|
| 1613 |
+
border-left-color: #718096 !important;
|
| 1614 |
+
}
|
| 1615 |
+
html[data-theme="dark"] .info-box p strong,
|
| 1616 |
+
.dark .info-box p strong {
|
| 1617 |
+
color: #f7fafc !important;
|
| 1618 |
+
}
|
| 1619 |
+
html[data-theme="dark"] .response-box,
|
| 1620 |
+
.dark .response-box {
|
| 1621 |
+
background-color: #1a202c !important;
|
| 1622 |
+
color: #e2e8f0 !important;
|
| 1623 |
+
border-left-color: #4a5568 !important;
|
| 1624 |
+
}
|
| 1625 |
+
html[data-theme="dark"] .example-query,
|
| 1626 |
+
.dark .example-query {
|
| 1627 |
+
background-color: #2a4365 !important;
|
| 1628 |
+
border-color: #2c5282 !important;
|
| 1629 |
+
color: #bee3f8 !important;
|
| 1630 |
+
}
|
| 1631 |
+
html[data-theme="dark"] .stats-section,
|
| 1632 |
+
.dark .stats-section {
|
| 1633 |
+
background-color: #2a4365 !important;
|
| 1634 |
+
border-color: #2c5282 !important;
|
| 1635 |
+
}
|
| 1636 |
+
html[data-theme="dark"] .stat-value,
|
| 1637 |
+
.dark .stat-value {
|
| 1638 |
+
color: #bee3f8 !important;
|
| 1639 |
+
}
|
| 1640 |
+
html[data-theme="dark"] .stat-label,
|
| 1641 |
+
.dark .stat-label {
|
| 1642 |
+
color: #90cdf4 !important;
|
| 1643 |
+
}
|
| 1644 |
+
html[data-theme="dark"] .processing-status,
|
| 1645 |
+
.dark .processing-status {
|
| 1646 |
+
background-color: #2a4365 !important;
|
| 1647 |
+
border-left-color: #90cdf4 !important;
|
| 1648 |
+
color: #bee3f8 !important;
|
| 1649 |
+
}
|
| 1650 |
+
html[data-theme="dark"] .debug-panel,
|
| 1651 |
+
.dark .debug-panel {
|
| 1652 |
+
background-color: #1a202c !important;
|
| 1653 |
+
border-color: #4a5568 !important;
|
| 1654 |
+
color: #e2e8f0 !important;
|
| 1655 |
+
}
|
| 1656 |
+
html[data-theme="dark"] .progress-container,
|
| 1657 |
+
.dark .progress-container {
|
| 1658 |
+
background-color: #2d3748 !important;
|
| 1659 |
+
border-color: #4a5568 !important;
|
| 1660 |
+
}
|
| 1661 |
+
html[data-theme="dark"] .progress-bar-container,
|
| 1662 |
+
.dark .progress-bar-container {
|
| 1663 |
+
background-color: #4a5568 !important;
|
| 1664 |
+
}
|
| 1665 |
+
html[data-theme="dark"] .query-display,
|
| 1666 |
+
.dark .query-display {
|
| 1667 |
+
color: #a0aec0 !important;
|
| 1668 |
+
background-color: #1a202c !important;
|
| 1669 |
+
border-left-color: #90cdf4 !important;
|
| 1670 |
+
}
|
| 1671 |
+
|
| 1672 |
+
/* Additional theme-aware classes */
|
| 1673 |
+
.model-info-bar {
|
| 1674 |
+
background-color: #ebf8ff;
|
| 1675 |
+
padding: 10px 15px;
|
| 1676 |
+
border-radius: 8px;
|
| 1677 |
+
margin-bottom: 15px;
|
| 1678 |
+
display: flex;
|
| 1679 |
+
justify-content: space-between;
|
| 1680 |
+
border: 1px solid #bee3f8;
|
| 1681 |
+
}
|
| 1682 |
+
.model-info-section {
|
| 1683 |
+
flex: 1;
|
| 1684 |
+
text-align: center;
|
| 1685 |
+
padding-right: 10px;
|
| 1686 |
+
border-right: 1px solid #bee3f8;
|
| 1687 |
+
}
|
| 1688 |
+
.model-info-section:last-child {
|
| 1689 |
+
border-right: none;
|
| 1690 |
+
padding-right: 0;
|
| 1691 |
+
padding-left: 10px;
|
| 1692 |
+
}
|
| 1693 |
+
.model-label {
|
| 1694 |
+
font-weight: bold;
|
| 1695 |
+
color: #2c5282;
|
| 1696 |
+
}
|
| 1697 |
+
.model-name {
|
| 1698 |
+
font-size: 1.2em;
|
| 1699 |
+
color: #2b6cb0;
|
| 1700 |
+
}
|
| 1701 |
+
.app-title {
|
| 1702 |
+
font-size: 2.2em;
|
| 1703 |
+
font-weight: 600;
|
| 1704 |
+
color: #2c5282;
|
| 1705 |
+
margin-bottom: 0.2em;
|
| 1706 |
+
}
|
| 1707 |
+
.app-subtitle {
|
| 1708 |
+
font-size: 1.3em;
|
| 1709 |
+
color: #4a5568;
|
| 1710 |
+
margin-bottom: 0.8em;
|
| 1711 |
+
}
|
| 1712 |
+
.app-description {
|
| 1713 |
+
font-size: 1.1em;
|
| 1714 |
+
color: #718096;
|
| 1715 |
+
max-width: 800px;
|
| 1716 |
+
margin: 0 auto;
|
| 1717 |
+
}
|
| 1718 |
+
.section-meta {
|
| 1719 |
+
font-size: 0.8em;
|
| 1720 |
+
color: #718096;
|
| 1721 |
+
}
|
| 1722 |
+
.divider-line {
|
| 1723 |
+
margin-top: 20px;
|
| 1724 |
+
border-top: 1px dashed #e2e8f0;
|
| 1725 |
+
padding-top: 15px;
|
| 1726 |
+
font-size: 0.9em;
|
| 1727 |
+
color: #718096;
|
| 1728 |
+
text-align: center;
|
| 1729 |
+
}
|
| 1730 |
+
.info-message {
|
| 1731 |
+
padding: 20px;
|
| 1732 |
+
background-color: #ebf8ff;
|
| 1733 |
+
border-radius: 8px;
|
| 1734 |
+
text-align: center;
|
| 1735 |
+
margin: 20px 0;
|
| 1736 |
+
border: 1px solid #bee3f8;
|
| 1737 |
+
}
|
| 1738 |
+
.info-message h3 {
|
| 1739 |
+
margin-top: 0;
|
| 1740 |
+
color: #2c5282;
|
| 1741 |
+
}
|
| 1742 |
+
.error-message {
|
| 1743 |
+
padding: 20px;
|
| 1744 |
+
background-color: #ffeaea;
|
| 1745 |
+
border-radius: 8px;
|
| 1746 |
+
text-align: center;
|
| 1747 |
+
margin: 20px 0;
|
| 1748 |
+
border: 1px solid #fc8181;
|
| 1749 |
+
}
|
| 1750 |
+
.error-message h3 {
|
| 1751 |
+
margin-top: 0;
|
| 1752 |
+
color: #e53e3e;
|
| 1753 |
+
}
|
| 1754 |
+
.perf-metric {
|
| 1755 |
+
font-weight: 500;
|
| 1756 |
+
}
|
| 1757 |
+
.perf-generator {
|
| 1758 |
+
color: #38a169;
|
| 1759 |
+
}
|
| 1760 |
+
.perf-judge {
|
| 1761 |
+
color: #3182ce;
|
| 1762 |
+
}
|
| 1763 |
+
.perf-consistency {
|
| 1764 |
+
color: #805ad5;
|
| 1765 |
+
}
|
| 1766 |
+
.perf-distribution {
|
| 1767 |
+
color: #d69e2e;
|
| 1768 |
+
}
|
| 1769 |
+
|
| 1770 |
+
/* Dark theme versions */
|
| 1771 |
+
@media (prefers-color-scheme: dark) {
|
| 1772 |
+
.model-info-bar {
|
| 1773 |
+
background-color: #2a4365;
|
| 1774 |
+
border-color: #2c5282;
|
| 1775 |
+
}
|
| 1776 |
+
.model-info-section {
|
| 1777 |
+
border-right-color: #2c5282;
|
| 1778 |
+
}
|
| 1779 |
+
.model-label {
|
| 1780 |
+
color: #bee3f8;
|
| 1781 |
+
}
|
| 1782 |
+
.model-name {
|
| 1783 |
+
color: #90cdf4;
|
| 1784 |
+
}
|
| 1785 |
+
.app-title {
|
| 1786 |
+
color: #63b3ed;
|
| 1787 |
+
}
|
| 1788 |
+
.app-subtitle {
|
| 1789 |
+
color: #a0aec0;
|
| 1790 |
+
}
|
| 1791 |
+
.app-description {
|
| 1792 |
+
color: #cbd5e0;
|
| 1793 |
+
}
|
| 1794 |
+
.section-meta {
|
| 1795 |
+
color: #a0aec0;
|
| 1796 |
+
}
|
| 1797 |
+
.divider-line {
|
| 1798 |
+
border-top-color: #4a5568;
|
| 1799 |
+
color: #a0aec0;
|
| 1800 |
+
}
|
| 1801 |
+
.info-message {
|
| 1802 |
+
background-color: #2a4365;
|
| 1803 |
+
border-color: #2c5282;
|
| 1804 |
+
}
|
| 1805 |
+
.info-message h3 {
|
| 1806 |
+
color: #bee3f8;
|
| 1807 |
+
}
|
| 1808 |
+
.error-message {
|
| 1809 |
+
background-color: #553c39;
|
| 1810 |
+
border-color: #fc8181;
|
| 1811 |
+
}
|
| 1812 |
+
.error-message h3 {
|
| 1813 |
+
color: #fc8181;
|
| 1814 |
+
}
|
| 1815 |
+
.perf-generator {
|
| 1816 |
+
color: #68d391;
|
| 1817 |
+
}
|
| 1818 |
+
.perf-judge {
|
| 1819 |
+
color: #90cdf4;
|
| 1820 |
+
}
|
| 1821 |
+
.perf-consistency {
|
| 1822 |
+
color: #b794f6;
|
| 1823 |
+
}
|
| 1824 |
+
.perf-distribution {
|
| 1825 |
+
color: #f6e05e;
|
| 1826 |
+
}
|
| 1827 |
+
}
|
| 1828 |
+
|
| 1829 |
+
/* Gradio fallbacks for new classes */
|
| 1830 |
+
html[data-theme="dark"] .model-info-bar,
|
| 1831 |
+
.dark .model-info-bar {
|
| 1832 |
+
background-color: #2a4365 !important;
|
| 1833 |
+
border-color: #2c5282 !important;
|
| 1834 |
+
}
|
| 1835 |
+
html[data-theme="dark"] .model-label,
|
| 1836 |
+
.dark .model-label {
|
| 1837 |
+
color: #bee3f8 !important;
|
| 1838 |
+
}
|
| 1839 |
+
html[data-theme="dark"] .model-name,
|
| 1840 |
+
.dark .model-name {
|
| 1841 |
+
color: #90cdf4 !important;
|
| 1842 |
+
}
|
| 1843 |
+
html[data-theme="dark"] .app-title,
|
| 1844 |
+
.dark .app-title {
|
| 1845 |
+
color: #63b3ed !important;
|
| 1846 |
+
}
|
| 1847 |
+
html[data-theme="dark"] .app-subtitle,
|
| 1848 |
+
.dark .app-subtitle {
|
| 1849 |
+
color: #a0aec0 !important;
|
| 1850 |
+
}
|
| 1851 |
+
html[data-theme="dark"] .app-description,
|
| 1852 |
+
.dark .app-description {
|
| 1853 |
+
color: #cbd5e0 !important;
|
| 1854 |
+
}
|
| 1855 |
+
html[data-theme="dark"] .section-meta,
|
| 1856 |
+
.dark .section-meta {
|
| 1857 |
+
color: #a0aec0 !important;
|
| 1858 |
+
}
|
| 1859 |
+
html[data-theme="dark"] .divider-line,
|
| 1860 |
+
.dark .divider-line {
|
| 1861 |
+
border-top-color: #4a5568 !important;
|
| 1862 |
+
color: #a0aec0 !important;
|
| 1863 |
+
}
|
| 1864 |
+
|
| 1865 |
+
/* Progress model info styling */
|
| 1866 |
+
.progress-model-info {
|
| 1867 |
+
display: flex;
|
| 1868 |
+
justify-content: space-between;
|
| 1869 |
+
margin-top: 8px;
|
| 1870 |
+
font-size: 0.85em;
|
| 1871 |
+
color: #4a5568;
|
| 1872 |
+
background-color: #ebf8ff;
|
| 1873 |
+
padding: 5px 10px;
|
| 1874 |
+
border-radius: 4px;
|
| 1875 |
+
border: 1px solid #bee3f8;
|
| 1876 |
+
}
|
| 1877 |
+
|
| 1878 |
+
@media (prefers-color-scheme: dark) {
|
| 1879 |
+
.progress-model-info {
|
| 1880 |
+
color: #a0aec0;
|
| 1881 |
+
background-color: #2a4365;
|
| 1882 |
+
border-color: #2c5282;
|
| 1883 |
+
}
|
| 1884 |
+
}
|
| 1885 |
+
|
| 1886 |
+
html[data-theme="dark"] .progress-model-info,
|
| 1887 |
+
.dark .progress-model-info {
|
| 1888 |
+
color: #a0aec0 !important;
|
| 1889 |
+
background-color: #2a4365 !important;
|
| 1890 |
+
border-color: #2c5282 !important;
|
| 1891 |
+
}
|
| 1892 |
+
|
| 1893 |
+
/* Metrics explanation box styling */
|
| 1894 |
+
.metrics-explanation {
|
| 1895 |
+
margin-top: 15px;
|
| 1896 |
+
padding: 12px;
|
| 1897 |
+
background-color: #f7fafc;
|
| 1898 |
+
border-radius: 8px;
|
| 1899 |
+
font-size: 0.95em;
|
| 1900 |
+
color: #2d3748;
|
| 1901 |
+
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
| 1902 |
+
border: 1px solid #e2e8f0;
|
| 1903 |
+
}
|
| 1904 |
+
.metrics-explanation p {
|
| 1905 |
+
margin-bottom: 8px;
|
| 1906 |
+
color: #2c5282;
|
| 1907 |
+
}
|
| 1908 |
+
.metrics-explanation ul {
|
| 1909 |
+
margin-top: 5px;
|
| 1910 |
+
padding-left: 20px;
|
| 1911 |
+
line-height: 1.4;
|
| 1912 |
+
}
|
| 1913 |
+
.metrics-explanation strong {
|
| 1914 |
+
color: #2b6cb0;
|
| 1915 |
+
}
|
| 1916 |
+
|
| 1917 |
+
@media (prefers-color-scheme: dark) {
|
| 1918 |
+
.metrics-explanation {
|
| 1919 |
+
background-color: #2d3748;
|
| 1920 |
+
color: #e2e8f0;
|
| 1921 |
+
border-color: #4a5568;
|
| 1922 |
+
}
|
| 1923 |
+
.metrics-explanation p {
|
| 1924 |
+
color: #90cdf4;
|
| 1925 |
+
}
|
| 1926 |
+
.metrics-explanation strong {
|
| 1927 |
+
color: #bee3f8;
|
| 1928 |
+
}
|
| 1929 |
+
}
|
| 1930 |
+
|
| 1931 |
+
html[data-theme="dark"] .metrics-explanation,
|
| 1932 |
+
.dark .metrics-explanation {
|
| 1933 |
+
background-color: #2d3748 !important;
|
| 1934 |
+
color: #e2e8f0 !important;
|
| 1935 |
+
border-color: #4a5568 !important;
|
| 1936 |
+
}
|
| 1937 |
+
html[data-theme="dark"] .metrics-explanation p,
|
| 1938 |
+
.dark .metrics-explanation p {
|
| 1939 |
+
color: #90cdf4 !important;
|
| 1940 |
+
}
|
| 1941 |
+
html[data-theme="dark"] .metrics-explanation strong,
|
| 1942 |
+
.dark .metrics-explanation strong {
|
| 1943 |
+
color: #bee3f8 !important;
|
| 1944 |
+
}
|
| 1945 |
+
|
| 1946 |
+
/* Leaderboard table styling */
|
| 1947 |
+
.leaderboard-container {
|
| 1948 |
+
margin: 15px 0;
|
| 1949 |
+
overflow-x: auto;
|
| 1950 |
+
}
|
| 1951 |
+
.leaderboard-table {
|
| 1952 |
+
width: 100%;
|
| 1953 |
+
border-collapse: collapse;
|
| 1954 |
+
font-size: 0.95em;
|
| 1955 |
+
box-shadow: 0 2px 10px rgba(0,0,0,0.1);
|
| 1956 |
+
border-radius: 8px;
|
| 1957 |
+
overflow: hidden;
|
| 1958 |
+
border: 1px solid #e2e8f0;
|
| 1959 |
+
}
|
| 1960 |
+
.leaderboard-table thead {
|
| 1961 |
+
background-color: #3182ce;
|
| 1962 |
+
color: white;
|
| 1963 |
+
}
|
| 1964 |
+
.leaderboard-table th,
|
| 1965 |
+
.leaderboard-table td {
|
| 1966 |
+
padding: 12px 15px;
|
| 1967 |
+
text-align: left;
|
| 1968 |
+
border-bottom: 1px solid #e2e8f0;
|
| 1969 |
+
color: #2d3748;
|
| 1970 |
+
}
|
| 1971 |
+
.leaderboard-table thead th {
|
| 1972 |
+
color: white;
|
| 1973 |
+
border-bottom-color: #2c5282;
|
| 1974 |
+
}
|
| 1975 |
+
.leaderboard-table tbody tr {
|
| 1976 |
+
transition: background-color 0.3s;
|
| 1977 |
+
background-color: #ffffff;
|
| 1978 |
+
}
|
| 1979 |
+
.leaderboard-table tbody tr:nth-child(even) {
|
| 1980 |
+
background-color: #f7fafc;
|
| 1981 |
+
}
|
| 1982 |
+
.leaderboard-table tbody tr:hover {
|
| 1983 |
+
background-color: #ebf8ff;
|
| 1984 |
+
}
|
| 1985 |
+
.leaderboard-table tbody tr.top-rank-1 {
|
| 1986 |
+
background-color: #f0fff4;
|
| 1987 |
+
color: #22543d;
|
| 1988 |
+
font-weight: bold;
|
| 1989 |
+
}
|
| 1990 |
+
.leaderboard-table tbody tr.top-rank-2 {
|
| 1991 |
+
background-color: #fefcbf;
|
| 1992 |
+
color: #744210;
|
| 1993 |
+
font-weight: 500;
|
| 1994 |
+
}
|
| 1995 |
+
.leaderboard-table tbody tr.top-rank-3 {
|
| 1996 |
+
background-color: #fed7cc;
|
| 1997 |
+
color: #7c2d12;
|
| 1998 |
+
font-weight: 500;
|
| 1999 |
+
}
|
| 2000 |
+
|
| 2001 |
+
/* Dark theme leaderboard */
|
| 2002 |
+
@media (prefers-color-scheme: dark) {
|
| 2003 |
+
.leaderboard-table {
|
| 2004 |
+
border-color: #4a5568;
|
| 2005 |
+
box-shadow: 0 2px 10px rgba(0,0,0,0.3);
|
| 2006 |
+
}
|
| 2007 |
+
.leaderboard-table thead {
|
| 2008 |
+
background-color: #2c5282;
|
| 2009 |
+
}
|
| 2010 |
+
.leaderboard-table th,
|
| 2011 |
+
.leaderboard-table td {
|
| 2012 |
+
border-bottom-color: #4a5568;
|
| 2013 |
+
color: #e2e8f0;
|
| 2014 |
+
}
|
| 2015 |
+
.leaderboard-table thead th {
|
| 2016 |
+
border-bottom-color: #1a365d;
|
| 2017 |
+
}
|
| 2018 |
+
.leaderboard-table tbody tr {
|
| 2019 |
+
background-color: #2d3748;
|
| 2020 |
+
}
|
| 2021 |
+
.leaderboard-table tbody tr:nth-child(even) {
|
| 2022 |
+
background-color: #1a202c;
|
| 2023 |
+
}
|
| 2024 |
+
.leaderboard-table tbody tr:hover {
|
| 2025 |
+
background-color: #2a4365;
|
| 2026 |
+
}
|
| 2027 |
+
.leaderboard-table tbody tr.top-rank-1 {
|
| 2028 |
+
background-color: #22543d;
|
| 2029 |
+
color: #c6f6d5;
|
| 2030 |
+
}
|
| 2031 |
+
.leaderboard-table tbody tr.top-rank-2 {
|
| 2032 |
+
background-color: #744210;
|
| 2033 |
+
color: #fefcbf;
|
| 2034 |
+
}
|
| 2035 |
+
.leaderboard-table tbody tr.top-rank-3 {
|
| 2036 |
+
background-color: #7c2d12;
|
| 2037 |
+
color: #fed7cc;
|
| 2038 |
+
}
|
| 2039 |
+
}
|
| 2040 |
+
|
| 2041 |
+
/* Gradio fallbacks for leaderboard */
|
| 2042 |
+
html[data-theme="dark"] .leaderboard-table,
|
| 2043 |
+
.dark .leaderboard-table {
|
| 2044 |
+
border-color: #4a5568 !important;
|
| 2045 |
+
box-shadow: 0 2px 10px rgba(0,0,0,0.3) !important;
|
| 2046 |
+
}
|
| 2047 |
+
html[data-theme="dark"] .leaderboard-table thead,
|
| 2048 |
+
.dark .leaderboard-table thead {
|
| 2049 |
+
background-color: #2c5282 !important;
|
| 2050 |
+
}
|
| 2051 |
+
html[data-theme="dark"] .leaderboard-table th,
|
| 2052 |
+
html[data-theme="dark"] .leaderboard-table td,
|
| 2053 |
+
.dark .leaderboard-table th,
|
| 2054 |
+
.dark .leaderboard-table td {
|
| 2055 |
+
border-bottom-color: #4a5568 !important;
|
| 2056 |
+
color: #e2e8f0 !important;
|
| 2057 |
+
}
|
| 2058 |
+
html[data-theme="dark"] .leaderboard-table thead th,
|
| 2059 |
+
.dark .leaderboard-table thead th {
|
| 2060 |
+
border-bottom-color: #1a365d !important;
|
| 2061 |
+
color: white !important;
|
| 2062 |
+
}
|
| 2063 |
+
html[data-theme="dark"] .leaderboard-table tbody tr,
|
| 2064 |
+
.dark .leaderboard-table tbody tr {
|
| 2065 |
+
background-color: #2d3748 !important;
|
| 2066 |
+
}
|
| 2067 |
+
html[data-theme="dark"] .leaderboard-table tbody tr:nth-child(even),
|
| 2068 |
+
.dark .leaderboard-table tbody tr:nth-child(even) {
|
| 2069 |
+
background-color: #1a202c !important;
|
| 2070 |
+
}
|
| 2071 |
+
html[data-theme="dark"] .leaderboard-table tbody tr:hover,
|
| 2072 |
+
.dark .leaderboard-table tbody tr:hover {
|
| 2073 |
+
background-color: #2a4365 !important;
|
| 2074 |
}
|
| 2075 |
"""
|
| 2076 |
|
|
|
|
| 2337 |
<div class="container">
|
| 2338 |
<h2 class="title">Hallucination Detection Results</h2>
|
| 2339 |
|
| 2340 |
+
<div class="model-info-bar">
|
| 2341 |
+
<div class="model-info-section">
|
| 2342 |
+
<div class="model-label">Generator Model</div>
|
| 2343 |
+
<div class="model-name">{generator_model}</div>
|
| 2344 |
</div>
|
| 2345 |
+
<div class="model-info-section">
|
| 2346 |
+
<div class="model-label">Judge Model</div>
|
| 2347 |
+
<div class="model-name">{judge_model}</div>
|
| 2348 |
</div>
|
| 2349 |
</div>
|
| 2350 |
|
|
|
|
| 2377 |
{original_query}
|
| 2378 |
</div>
|
| 2379 |
|
| 2380 |
+
<div class="section-title">Original Response <span class="section-meta">(generated by {generator_model})</span></div>
|
| 2381 |
<div class="response-box">
|
| 2382 |
{original_response_safe}
|
| 2383 |
</div>
|
|
|
|
| 2392 |
{q}
|
| 2393 |
</div>
|
| 2394 |
|
| 2395 |
+
<div class="section-title">Response {i} <span class="section-meta">(generated by {generator_model})</span></div>
|
| 2396 |
<div class="response-box">
|
| 2397 |
{r}
|
| 2398 |
</div>
|
| 2399 |
"""
|
| 2400 |
|
| 2401 |
html_output += f"""
|
| 2402 |
+
<div class="section-title">Detailed Analysis <span class="section-meta">(judged by {judge_model})</span></div>
|
| 2403 |
<div class="info-box">
|
| 2404 |
<p><strong>Reasoning:</strong></p>
|
| 2405 |
<p>{reasoning_safe}</p>
|
|
|
|
| 2408 |
<p>{conflicting_facts_text_safe}</p>
|
| 2409 |
</div>
|
| 2410 |
|
| 2411 |
+
<div class="divider-line">
|
| 2412 |
Models randomly selected for this analysis: <strong>{generator_model}</strong> (Generator) and <strong>{judge_model}</strong> (Judge)
|
| 2413 |
</div>
|
| 2414 |
</div>
|
|
|
|
| 2533 |
gr.HTML(
|
| 2534 |
"""
|
| 2535 |
<div style="text-align: center; margin-bottom: 1.5rem">
|
| 2536 |
+
<h1 class="app-title">PAS2 - Hallucination Detector</h1>
|
| 2537 |
+
<h3 class="app-subtitle">Advanced AI Response Verification Using Model-as-Judge</h3>
|
| 2538 |
+
<p class="app-description">
|
| 2539 |
This tool detects hallucinations in AI responses by comparing answers to semantically equivalent questions and using a specialized judge model.
|
| 2540 |
</p>
|
| 2541 |
</div>
|
|
|
|
| 2663 |
|
| 2664 |
if not pairs:
|
| 2665 |
return (
|
| 2666 |
+
"<div class=\"info-message\">"
|
| 2667 |
+
"<h3>No Data Available Yet</h3>"
|
|
|
|
| 2668 |
"<p>Try the detector with more queries to populate the leaderboard!</p>"
|
| 2669 |
"</div>"
|
| 2670 |
)
|
|
|
|
| 2693 |
f"<td>{pair.get('judge', 'unknown')}</td>"
|
| 2694 |
f"<td>{round(pair.get('elo_score', 0))}</td>"
|
| 2695 |
f"<td>{pair.get('accuracy')}%</td>"
|
| 2696 |
+
f"<td class='perf-metric perf-generator'>{generator_perf}</td>"
|
| 2697 |
+
f"<td class='perf-metric perf-judge'>{judge_perf}</td>"
|
| 2698 |
+
f"<td class='perf-metric perf-consistency'>{consistency}</td>"
|
| 2699 |
f"<td>{pair.get('total_samples', 0)}</td>"
|
| 2700 |
f"</tr>"
|
| 2701 |
)
|
|
|
|
| 2722 |
f"</tbody>"
|
| 2723 |
f"</table>"
|
| 2724 |
f"</div>"
|
| 2725 |
+
f"<div class='metrics-explanation'>"
|
| 2726 |
+
f"<p><strong>Model Pair Performance Metrics:</strong></p>"
|
| 2727 |
+
f"<ul>"
|
| 2728 |
+
f"<li><strong>Accuracy</strong>: Percentage of correct hallucination judgments based on user feedback</li>"
|
| 2729 |
+
f"<li><strong>Generator Performance</strong>: How well the generator model avoids hallucinations</li>"
|
| 2730 |
+
f"<li><strong>Judge Performance</strong>: How accurately the judge model identifies hallucinations</li>"
|
| 2731 |
+
f"<li><strong>Consistency</strong>: Weighted measure of how well the pair works together</li>"
|
| 2732 |
f"</ul>"
|
| 2733 |
f"</div>"
|
| 2734 |
)
|
|
|
|
| 2737 |
except Exception as e:
|
| 2738 |
logger.error("Error generating leaderboard HTML: %s", str(e), exc_info=True)
|
| 2739 |
return (
|
| 2740 |
+
f"<div class=\"error-message\">"
|
| 2741 |
+
f"<h3>Error Loading Leaderboard</h3>"
|
|
|
|
| 2742 |
f"<p>{str(e)}</p>"
|
| 2743 |
f"</div>"
|
| 2744 |
)
|
|
|
|
| 2778 |
"<li>deepseek-reasoner</li><li>o4-mini</li><li>gemini-2.5-pro</li>" +
|
| 2779 |
"</ul></div></div></div></div></div>"
|
| 2780 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2781 |
|
| 2782 |
# Tab 3: Individual Models Leaderboard
|
| 2783 |
with gr.TabItem("Individual Models", elem_id="user-feedback-tab"):
|
|
|
|
| 2792 |
|
| 2793 |
if not models:
|
| 2794 |
return (
|
| 2795 |
+
"<div class=\"info-message\">"
|
| 2796 |
+
"<h3>No Data Available Yet</h3>"
|
|
|
|
| 2797 |
"<p>Try the detector with more queries to populate the model scores!</p>"
|
| 2798 |
"</div>"
|
| 2799 |
)
|
|
|
|
| 2830 |
f"<td>{model.get('model_name', 'unknown')}</td>"
|
| 2831 |
f"<td>{round(model.get('elo_score', 0))}</td>"
|
| 2832 |
f"<td>{model.get('accuracy')}%</td>"
|
| 2833 |
+
f"<td class='perf-metric perf-generator'>{generator_acc}</td>"
|
| 2834 |
+
f"<td class='perf-metric perf-judge'>{judge_acc}</td>"
|
| 2835 |
f"<td>{model.get('total_samples', 0)}</td>"
|
| 2836 |
+
f"<td class='perf-metric perf-distribution'>{role_distribution}</td>"
|
| 2837 |
f"</tr>"
|
| 2838 |
)
|
| 2839 |
|
|
|
|
| 2864 |
except Exception as e:
|
| 2865 |
logger.error("Error generating model leaderboard HTML: %s", str(e), exc_info=True)
|
| 2866 |
return (
|
| 2867 |
+
f"<div class=\"error-message\">"
|
| 2868 |
+
f"<h3>Error Loading Model Leaderboard</h3>"
|
|
|
|
| 2869 |
f"<p>{str(e)}</p>"
|
| 2870 |
f"</div>"
|
| 2871 |
)
|
|
|
|
| 2922 |
accuracy_pct = f"{accuracy * 100:.1f}%"
|
| 2923 |
|
| 2924 |
stats_html = f"""
|
| 2925 |
+
<div class="stats-section">
|
| 2926 |
<div class="stat-item">
|
| 2927 |
+
<div class="stat-value">{total}</div>
|
| 2928 |
+
<div class="stat-label">Total Responses</div>
|
| 2929 |
</div>
|
| 2930 |
<div class="stat-item">
|
| 2931 |
+
<div class="stat-value">{accuracy_pct}</div>
|
| 2932 |
+
<div class="stat-label">Correct Predictions</div>
|
| 2933 |
</div>
|
| 2934 |
</div>
|
| 2935 |
+
<div class="section-meta" style="text-align: center; margin-top: 10px; font-style: italic;">
|
| 2936 |
Based on user feedback: {correct} correct out of {total} total predictions
|
| 2937 |
</div>
|
| 2938 |
"""
|