FutureBench / src /display_utils.py
vinid's picture
Leaderboard deployment 2025-07-16 18:05:41
6441bc6
from dataclasses import dataclass
from .about import Tasks
@dataclass(frozen=True)
class ColumnContent:
name: str
type: str
displayed_by_default: bool
hidden: bool = False
# Define leaderboard columns
@dataclass(frozen=True)
class LeaderboardColumn:
model = ColumnContent("Model", "str", True)
events = ColumnContent("Events", "number", True)
average = ColumnContent("Average", "number", True)
# Task-specific columns will be added dynamically
# Additional model info (hidden by default)
correct_predictions = ColumnContent("Correct Predictions", "number", False)
# Get column names for display
def get_display_columns():
"""Get list of column names for display"""
base_cols = ["Rank", "Model", "Events", "Average"]
task_cols = [task.value.col_name for task in Tasks]
return base_cols + task_cols
def get_all_columns():
"""Get all column names including hidden ones"""
base_cols = get_display_columns()
hidden_cols = ["Correct Predictions"]
return base_cols + hidden_cols
# Formatting helpers
def make_clickable_model(model_name):
"""Make model name clickable with link to HuggingFace"""
if "/" in model_name:
link = f"https://huggingface.co/{model_name}"
return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;">{model_name}</a>'
return model_name
def format_percentage(value):
"""Format accuracy as percentage"""
if value is None or value == "N/A":
return "N/A"
try:
return f"{float(value):.1f}%"
except (ValueError, TypeError):
return "N/A"
def has_valid_scores(df, required_columns):
"""Check if dataframe has valid scores for required columns"""
return df[required_columns].notna().all(axis=1)
# CSS styling
CUSTOM_CSS = """
/* Global styling */
body {
background: linear-gradient(135deg, #1e1e2f 0%, #2d2d44 100%) !important;
}
/* Add consistent margins and centering */
.gradio-container,
.container,
.main {
margin: 0 auto !important;
max-width: 1400px !important;
padding: 0 60px !important;
}
.block {
margin: 0 auto !important;
max-width: 100% !important;
}
.markdown-text {
font-size: 18px !important;
line-height: 1.6 !important;
}
/* Larger font for introduction text */
.section-card {
font-size: 22px !important;
line-height: 1.7 !important;
}
.section-card p {
font-size: 22px !important;
line-height: 1.7 !important;
}
.section-card .markdown-text {
font-size: 22px !important;
line-height: 1.7 !important;
}
/* Header styling */
#space-title {
text-shadow: 2px 2px 4px rgba(0,0,0,0.3) !important;
margin-bottom: 0.5rem !important;
}
.center-logo {
display: flex !important;
justify-content: center !important;
align-items: center !important;
margin: 0.25rem 0 0.5rem 0 !important;
}
.center-logo img {
width: 200px !important;
height: 200px !important;
border-radius: 50% !important;
overflow: hidden !important;
object-fit: cover !important;
box-shadow: 0 8px 32px rgba(0,0,0,0.3) !important;
border: 3px solid rgba(255,255,255,0.1) !important;
}
/* Tab styling */
.tab-nav {
margin: 1rem 0 !important;
display: flex !important;
justify-content: center !important;
}
.tab-buttons {
display: flex !important;
justify-content: center !important;
flex-wrap: wrap !important;
gap: 8px !important;
}
.tab-buttons button {
font-size: 22px !important;
padding: 16px 32px !important;
margin: 0 6px !important;
border-radius: 8px !important;
border: 2px solid transparent !important;
background: rgba(255,255,255,0.1) !important;
color: white !important;
transition: all 0.3s ease !important;
}
.tab-buttons button:hover {
background: rgba(255,255,255,0.2) !important;
transform: translateY(-2px) !important;
}
.tab-buttons button.selected {
background: linear-gradient(135deg, #6366f1, #8b5cf6) !important;
border-color: #6366f1 !important;
box-shadow: 0 4px 12px rgba(99, 102, 241, 0.3) !important;
}
/* Leaderboard table styling */
#leaderboard-table {
margin: 20px 0 !important;
border-radius: 12px !important;
overflow: hidden !important;
box-shadow: 0 8px 32px rgba(0,0,0,0.2) !important;
}
#leaderboard-table table {
border-collapse: separate !important;
border-spacing: 0 !important;
width: 100% !important;
}
#leaderboard-table th {
background: linear-gradient(135deg, #4f46e5, #6366f1) !important;
color: white !important;
padding: 22px !important;
font-weight: 600 !important;
text-align: left !important;
border: none !important;
font-size: 16px !important;
}
#leaderboard-table td {
padding: 20px 22px !important;
border: none !important;
font-size: 16px !important;
}
#leaderboard-table tr:nth-child(even) {
background: rgba(255,255,255,0.05) !important;
}
#leaderboard-table tr:hover {
background: rgba(99, 102, 241, 0.1) !important;
transform: scale(1.01) !important;
transition: all 0.2s ease !important;
}
/* Rank column styling */
#leaderboard-table td:nth-child(1),
#leaderboard-table th:nth-child(1) {
text-align: center !important;
width: 80px !important;
min-width: 80px !important;
max-width: 80px !important;
font-size: 18px !important;
font-weight: 600 !important;
}
/* Model column styling */
#leaderboard-table td:nth-child(2),
#leaderboard-table th:nth-child(2) {
min-width: 180px !important;
max-width: 300px !important;
overflow: hidden !important;
white-space: nowrap !important;
text-overflow: ellipsis !important;
font-size: 16px !important;
}
/* Events column styling (numeric) */
#leaderboard-table td:nth-child(3),
#leaderboard-table th:nth-child(3) {
text-align: center !important;
width: 90px !important;
min-width: 90px !important;
max-width: 90px !important;
font-size: 16px !important;
font-weight: 600 !important;
}
/* Average column styling (percentage) */
#leaderboard-table td:nth-child(4),
#leaderboard-table th:nth-child(4) {
text-align: center !important;
width: 110px !important;
min-width: 110px !important;
max-width: 110px !important;
font-size: 17px !important;
font-weight: 700 !important;
color: #10b981 !important;
}
/* Task-specific columns (News, PolyMarket) - compact percentage columns */
#leaderboard-table td:nth-child(n+5),
#leaderboard-table th:nth-child(n+5) {
text-align: center !important;
width: 100px !important;
min-width: 100px !important;
max-width: 100px !important;
font-size: 16px !important;
font-weight: 600 !important;
}
/* Dropdown styling */
.dropdown {
margin: 20px 0 !important;
width: 100% !important;
}
.dropdown select {
background: rgba(255,255,255,0.1) !important;
border: 2px solid rgba(255,255,255,0.2) !important;
border-radius: 8px !important;
padding: 12px 18px !important;
color: white !important;
font-size: 16px !important;
width: 100% !important;
max-width: 300px !important;
}
/* Button styling */
#refresh-button, .refresh-btn {
background: linear-gradient(135deg, #10b981, #059669) !important;
color: white !important;
border: none !important;
padding: 14px 28px !important;
border-radius: 8px !important;
cursor: pointer !important;
font-size: 18px !important;
font-weight: 500 !important;
transition: all 0.3s ease !important;
box-shadow: 0 4px 12px rgba(16, 185, 129, 0.3) !important;
}
#refresh-button:hover, .refresh-btn:hover {
background: linear-gradient(135deg, #059669, #047857) !important;
transform: translateY(-2px) !important;
box-shadow: 0 6px 16px rgba(16, 185, 129, 0.4) !important;
}
/* Cards and sections */
.section-card {
background: rgba(255,255,255,0.05) !important;
border-radius: 12px !important;
padding: 25px !important;
margin: 15px 0 !important;
border: 1px solid rgba(255,255,255,0.1) !important;
box-shadow: 0 4px 16px rgba(0,0,0,0.1) !important;
max-width: 100% !important;
}
/* Metrics and stats */
.metric-highlight {
color: #10b981 !important;
font-weight: 600 !important;
}
.model-rank-1 {
background: linear-gradient(135deg, #fbbf24, #f59e0b) !important;
color: #1f2937 !important;
font-weight: 600 !important;
}
.model-rank-2 {
background: linear-gradient(135deg, #e5e7eb, #d1d5db) !important;
color: #1f2937 !important;
font-weight: 600 !important;
}
.model-rank-3 {
background: linear-gradient(135deg, #cd7c2f, #a16207) !important;
color: white !important;
font-weight: 600 !important;
}
/* Performance badges */
.rank-badge {
display: inline-block !important;
padding: 4px 8px !important;
border-radius: 20px !important;
font-size: 10px !important;
font-weight: 600 !important;
margin-right: 8px !important;
}
.rank-1 .rank-badge {
background: linear-gradient(135deg, #fbbf24, #f59e0b) !important;
color: #1f2937 !important;
}
.rank-2 .rank-badge {
background: linear-gradient(135deg, #e5e7eb, #d1d5db) !important;
color: #1f2937 !important;
}
.rank-3 .rank-badge {
background: linear-gradient(135deg, #cd7c2f, #a16207) !important;
color: white !important;
}
/* Progress bars for accuracy */
.accuracy-bar {
width: 100% !important;
height: 6px !important;
background: rgba(255,255,255,0.1) !important;
border-radius: 3px !important;
margin-top: 4px !important;
overflow: hidden !important;
}
.accuracy-progress {
height: 100% !important;
background: linear-gradient(90deg, #10b981, #059669) !important;
border-radius: 3px !important;
transition: width 0.8s ease !important;
}
/* Enhanced summary section */
.summary-stats {
display: grid !important;
grid-template-columns: repeat(auto-fit, minmax(200px, 1fr)) !important;
gap: 20px !important;
margin: 20px 0 !important;
}
.stat-card {
background: rgba(255,255,255,0.08) !important;
border-radius: 12px !important;
padding: 20px !important;
border: 1px solid rgba(255,255,255,0.1) !important;
text-align: center !important;
transition: transform 0.3s ease !important;
}
.stat-card:hover {
transform: translateY(-4px) !important;
}
.stat-value {
font-size: 1.875rem !important;
font-weight: 700 !important;
color: #10b981 !important;
margin-bottom: 8px !important;
}
.stat-label {
font-size: 0.775rem !important;
color: rgba(255,255,255,0.7) !important;
text-transform: uppercase !important;
letter-spacing: 0.5px !important;
}
/* Better section headers */
.section-header {
display: flex !important;
align-items: center !important;
gap: 12px !important;
margin: 0 0 15px 0 !important;
font-size: 1.675rem !important;
font-weight: 600 !important;
}
.section-icon {
font-size: 1.375rem !important;
}
/* Improved table styling */
#leaderboard-table tr:first-child td:first-child {
position: relative !important;
}
#leaderboard-table tr:nth-child(1) {
background: rgba(251, 191, 36, 0.1) !important;
}
#leaderboard-table tr:nth-child(2) {
background: rgba(229, 231, 235, 0.1) !important;
}
#leaderboard-table tr:nth-child(3) {
background: rgba(205, 124, 47, 0.1) !important;
}
/* Loading animations */
@keyframes fadeInUp {
from {
opacity: 0;
transform: translateY(20px);
}
to {
opacity: 1;
transform: translateY(0);
}
}
.fade-in-up {
animation: fadeInUp 0.6s ease-out !important;
}
/* Staggered animations */
.stagger-1 { animation-delay: 0.1s !important; }
.stagger-2 { animation-delay: 0.2s !important; }
.stagger-3 { animation-delay: 0.3s !important; }
.stagger-4 { animation-delay: 0.4s !important; }
/* Enhanced buttons */
.icon-button {
display: inline-flex !important;
align-items: center !important;
gap: 8px !important;
}
.icon-button::before {
font-size: 1.0em !important;
}
/* Improved markdown styling */
.markdown-text h1 {
color: #10b981 !important;
border-bottom: 2px solid rgba(16, 185, 129, 0.3) !important;
padding-bottom: 8px !important;
}
.markdown-text h2 {
color: #6366f1 !important;
margin-top: 2rem !important;
}
.markdown-text h3 {
color: #8b5cf6 !important;
}
.markdown-text ul {
padding-left: 20px !important;
}
.markdown-text li {
margin: 8px 0 !important;
list-style-type: none !important;
position: relative !important;
}
.markdown-text li::before {
content: "▸" !important;
color: #10b981 !important;
position: absolute !important;
left: -16px !important;
font-weight: bold !important;
}
/* Responsive design */
@media (max-width: 768px) {
/* Adjust container margins for mobile */
.gradio-container,
.container,
.main {
padding: 0 30px !important;
}
#space-title {
font-size: 2.375rem !important;
}
.center-logo img {
width: 150px !important;
height: 150px !important;
}
.tab-buttons button {
font-size: 18px !important;
padding: 14px 24px !important;
}
.summary-stats {
grid-template-columns: 1fr !important;
}
.stat-value {
font-size: 1.375rem !important;
}
/* Maintain readable font sizes on mobile */
#leaderboard-table th {
font-size: 14px !important;
padding: 16px 12px !important;
}
#leaderboard-table td {
font-size: 14px !important;
padding: 16px 12px !important;
}
/* Adjust column widths for mobile */
#leaderboard-table td:nth-child(1),
#leaderboard-table th:nth-child(1) {
width: 60px !important;
min-width: 60px !important;
max-width: 60px !important;
}
#leaderboard-table td:nth-child(n+5),
#leaderboard-table th:nth-child(n+5) {
width: 90px !important;
min-width: 90px !important;
max-width: 90px !important;
}
}
"""