Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -71,7 +71,6 @@ custom_css += font_size_css
|
|
71 |
|
72 |
dark_css = """
|
73 |
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');
|
74 |
-
@import url('https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css');
|
75 |
|
76 |
body {
|
77 |
font-family: 'Inter', sans-serif;
|
@@ -93,6 +92,13 @@ body {
|
|
93 |
margin: 0;
|
94 |
font-size: 2.5rem !important;
|
95 |
font-weight: 700 !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
}
|
97 |
.hero-section h2 {
|
98 |
margin: .5rem 0 0 !important;
|
@@ -102,85 +108,183 @@ body {
|
|
102 |
}
|
103 |
|
104 |
/* Tab Buttons */
|
105 |
-
.tab-buttons button {
|
106 |
border-radius: 20px !important;
|
107 |
-
padding: 0.
|
108 |
-
margin-right: 0.
|
109 |
background: #1e1e1e !important;
|
110 |
color: #e0e0e0 !important;
|
111 |
border: none !important;
|
112 |
-
transition:
|
113 |
-
font-weight:
|
|
|
|
|
|
|
114 |
}
|
115 |
-
.tab-buttons button:hover {
|
116 |
background: #2c2c2c !important;
|
|
|
|
|
117 |
}
|
118 |
-
.tab-buttons button[aria-selected="true"] {
|
119 |
-
background: #444 !important;
|
120 |
color: #fff !important;
|
|
|
121 |
}
|
122 |
|
123 |
-
/* Category Selector
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
#category-selector input[type="radio"] { display: none; }
|
|
|
|
|
|
|
|
|
|
|
125 |
#category-selector label {
|
126 |
display: inline-block;
|
127 |
-
padding: 0.
|
128 |
-
margin-right: 0.5rem;
|
129 |
border-radius: 999px;
|
130 |
background: #1e1e1e;
|
131 |
cursor: pointer;
|
132 |
-
transition:
|
133 |
font-weight: 500;
|
134 |
color: #e0e0e0;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
}
|
136 |
#category-selector input[type="radio"]:checked + label {
|
137 |
-
background: #444;
|
138 |
color: #fff;
|
|
|
|
|
139 |
}
|
140 |
|
141 |
/* Table Styling */
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
.dataframe-container table {
|
143 |
width: 100%;
|
144 |
border: none;
|
145 |
-
|
146 |
-
border-radius: 0.5rem;
|
147 |
-
overflow: hidden;
|
148 |
}
|
149 |
.dataframe-container table th {
|
150 |
-
background: #2c2c2c;
|
151 |
-
color: #
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
}
|
153 |
.dataframe-container table tr:nth-child(odd) {
|
154 |
-
background-color: #
|
155 |
}
|
156 |
.dataframe-container table tr:nth-child(even) {
|
157 |
-
background-color: #
|
|
|
|
|
|
|
|
|
158 |
}
|
159 |
-
.dataframe-container table td
|
160 |
-
padding: 0.
|
161 |
color: #e0e0e0;
|
|
|
|
|
|
|
162 |
}
|
163 |
.dataframe-container table td a {
|
164 |
color: #8ab4f8;
|
165 |
text-decoration: none;
|
|
|
|
|
166 |
}
|
167 |
.dataframe-container table td a:hover {
|
168 |
color: #a3c9ff;
|
169 |
text-decoration: underline;
|
170 |
}
|
171 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
172 |
/* Enable scrollbar */
|
173 |
#leaderboard-table .dataframe-container {
|
174 |
-
max-height:
|
175 |
overflow-y: auto !important;
|
176 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
"""
|
178 |
custom_css += dark_css
|
179 |
|
180 |
# --- Override Title ---
|
181 |
TITLE = """
|
182 |
-
<div class
|
183 |
-
<h1><i class
|
184 |
<h2>Improving LLM Agents for Machine Learning Engineering</h2>
|
185 |
</div>
|
186 |
"""
|
@@ -190,24 +294,27 @@ TITLE = """
|
|
190 |
demo = gr.Blocks(css=custom_css, theme=gr.themes.Base())
|
191 |
|
192 |
with demo:
|
193 |
-
# Inject FontAwesome
|
194 |
gr.HTML("""
|
195 |
-
<
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
crossorigin
|
200 |
-
|
201 |
-
|
|
|
|
|
|
|
202 |
|
203 |
# Header & Intro
|
204 |
gr.HTML(TITLE)
|
205 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
206 |
|
207 |
-
# Tabs
|
208 |
-
with gr.Tabs(elem_classes="tab-buttons") as tabs:
|
209 |
-
with gr.TabItem("
|
210 |
-
gr.Markdown("## Model Elo Rankings by Category")
|
211 |
category_selector = gr.Radio(
|
212 |
choices=CATEGORIES,
|
213 |
label="Select Category:",
|
@@ -231,7 +338,7 @@ with demo:
|
|
231 |
outputs=leaderboard_df
|
232 |
)
|
233 |
|
234 |
-
with gr.TabItem("
|
235 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
236 |
|
237 |
# Citation Accordion
|
@@ -246,4 +353,4 @@ with demo:
|
|
246 |
|
247 |
if __name__ == "__main__":
|
248 |
print("Launching Gradio App in Dark Mode...")
|
249 |
-
demo.launch()
|
|
|
71 |
|
72 |
dark_css = """
|
73 |
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');
|
|
|
74 |
|
75 |
body {
|
76 |
font-family: 'Inter', sans-serif;
|
|
|
92 |
margin: 0;
|
93 |
font-size: 2.5rem !important;
|
94 |
font-weight: 700 !important;
|
95 |
+
display: flex;
|
96 |
+
align-items: center;
|
97 |
+
justify-content: center;
|
98 |
+
}
|
99 |
+
.hero-section h1 i {
|
100 |
+
margin-right: 0.5rem;
|
101 |
+
color: gold;
|
102 |
}
|
103 |
.hero-section h2 {
|
104 |
margin: .5rem 0 0 !important;
|
|
|
108 |
}
|
109 |
|
110 |
/* Tab Buttons */
|
111 |
+
.tab-nav-buttons button {
|
112 |
border-radius: 20px !important;
|
113 |
+
padding: 0.75rem 1.25rem !important;
|
114 |
+
margin-right: 0.75rem !important;
|
115 |
background: #1e1e1e !important;
|
116 |
color: #e0e0e0 !important;
|
117 |
border: none !important;
|
118 |
+
transition: all 0.3s ease !important;
|
119 |
+
font-weight: 600 !important;
|
120 |
+
font-size: 1.05rem !important;
|
121 |
+
letter-spacing: 0.02em !important;
|
122 |
+
box-shadow: 0 2px 4px rgba(0,0,0,0.2) !important;
|
123 |
}
|
124 |
+
.tab-nav-buttons button:hover {
|
125 |
background: #2c2c2c !important;
|
126 |
+
transform: translateY(-2px) !important;
|
127 |
+
box-shadow: 0 4px 8px rgba(0,0,0,0.3) !important;
|
128 |
}
|
129 |
+
.tab-nav-buttons button[aria-selected="true"] {
|
130 |
+
background: linear-gradient(to bottom, #444, #333) !important;
|
131 |
color: #fff !important;
|
132 |
+
box-shadow: 0 2px 6px rgba(0,0,0,0.4) !important;
|
133 |
}
|
134 |
|
135 |
+
/* Category Selector */
|
136 |
+
#category-selector .wrap {
|
137 |
+
display: flex;
|
138 |
+
flex-direction: column;
|
139 |
+
gap: 0.75rem;
|
140 |
+
}
|
141 |
+
#category-selector .label-wrap {
|
142 |
+
font-size: 1.15rem !important;
|
143 |
+
font-weight: 600 !important;
|
144 |
+
color: #e0e0e0 !important;
|
145 |
+
letter-spacing: 0.02em !important;
|
146 |
+
margin-bottom: 0.25rem !important;
|
147 |
+
}
|
148 |
#category-selector input[type="radio"] { display: none; }
|
149 |
+
#category-selector .options {
|
150 |
+
display: flex;
|
151 |
+
flex-wrap: wrap;
|
152 |
+
gap: 0.75rem;
|
153 |
+
}
|
154 |
#category-selector label {
|
155 |
display: inline-block;
|
156 |
+
padding: 0.65rem 1.25rem;
|
|
|
157 |
border-radius: 999px;
|
158 |
background: #1e1e1e;
|
159 |
cursor: pointer;
|
160 |
+
transition: all 0.3s ease;
|
161 |
font-weight: 500;
|
162 |
color: #e0e0e0;
|
163 |
+
letter-spacing: 0.02em;
|
164 |
+
font-size: 1.05rem;
|
165 |
+
box-shadow: 0 2px 4px rgba(0,0,0,0.15);
|
166 |
+
}
|
167 |
+
#category-selector label:hover {
|
168 |
+
background: #2a2a2a;
|
169 |
+
transform: translateY(-2px);
|
170 |
+
box-shadow: 0 4px 8px rgba(0,0,0,0.2);
|
171 |
}
|
172 |
#category-selector input[type="radio"]:checked + label {
|
173 |
+
background: linear-gradient(to bottom, #444, #333);
|
174 |
color: #fff;
|
175 |
+
box-shadow: 0 2px 6px rgba(0,0,0,0.3);
|
176 |
+
font-weight: 600;
|
177 |
}
|
178 |
|
179 |
/* Table Styling */
|
180 |
+
.dataframe-container {
|
181 |
+
margin-top: 1.5rem;
|
182 |
+
border-radius: 0.75rem;
|
183 |
+
overflow: hidden;
|
184 |
+
box-shadow: 0 4px 12px rgba(0,0,0,0.3);
|
185 |
+
}
|
186 |
.dataframe-container table {
|
187 |
width: 100%;
|
188 |
border: none;
|
189 |
+
border-collapse: collapse;
|
|
|
|
|
190 |
}
|
191 |
.dataframe-container table th {
|
192 |
+
background: linear-gradient(to bottom, #333, #2c2c2c);
|
193 |
+
color: #fff;
|
194 |
+
font-weight: 600;
|
195 |
+
font-size: 1.1rem;
|
196 |
+
text-align: left;
|
197 |
+
letter-spacing: 0.02em;
|
198 |
+
padding: 1rem 1.25rem;
|
199 |
+
position: sticky;
|
200 |
+
top: 0;
|
201 |
+
box-shadow: 0 2px 4px rgba(0,0,0,0.2);
|
202 |
+
z-index: 10;
|
203 |
}
|
204 |
.dataframe-container table tr:nth-child(odd) {
|
205 |
+
background-color: #1a1a1a !important;
|
206 |
}
|
207 |
.dataframe-container table tr:nth-child(even) {
|
208 |
+
background-color: #222 !important;
|
209 |
+
}
|
210 |
+
.dataframe-container table tr:hover {
|
211 |
+
background-color: #2c2c2c !important;
|
212 |
+
transition: background-color 0.2s ease;
|
213 |
}
|
214 |
+
.dataframe-container table td {
|
215 |
+
padding: 0.9rem 1.25rem;
|
216 |
color: #e0e0e0;
|
217 |
+
font-size: 1.05rem;
|
218 |
+
letter-spacing: 0.01em;
|
219 |
+
border-bottom: 1px solid #333;
|
220 |
}
|
221 |
.dataframe-container table td a {
|
222 |
color: #8ab4f8;
|
223 |
text-decoration: none;
|
224 |
+
font-weight: 500;
|
225 |
+
transition: all 0.2s ease;
|
226 |
}
|
227 |
.dataframe-container table td a:hover {
|
228 |
color: #a3c9ff;
|
229 |
text-decoration: underline;
|
230 |
}
|
231 |
|
232 |
+
/* Column-specific styles */
|
233 |
+
.dataframe-container table td:first-child,
|
234 |
+
.dataframe-container table th:first-child {
|
235 |
+
text-align: center;
|
236 |
+
width: 80px;
|
237 |
+
}
|
238 |
+
.dataframe-container table td:last-child,
|
239 |
+
.dataframe-container table th:last-child {
|
240 |
+
text-align: center;
|
241 |
+
font-weight: 600;
|
242 |
+
}
|
243 |
+
|
244 |
/* Enable scrollbar */
|
245 |
#leaderboard-table .dataframe-container {
|
246 |
+
max-height: 500px !important;
|
247 |
overflow-y: auto !important;
|
248 |
}
|
249 |
+
|
250 |
+
/* Custom Scrollbar for WebKit browsers */
|
251 |
+
.dataframe-container::-webkit-scrollbar {
|
252 |
+
width: 8px;
|
253 |
+
}
|
254 |
+
.dataframe-container::-webkit-scrollbar-track {
|
255 |
+
background: #1a1a1a;
|
256 |
+
border-radius: 4px;
|
257 |
+
}
|
258 |
+
.dataframe-container::-webkit-scrollbar-thumb {
|
259 |
+
background: #444;
|
260 |
+
border-radius: 4px;
|
261 |
+
}
|
262 |
+
.dataframe-container::-webkit-scrollbar-thumb:hover {
|
263 |
+
background: #555;
|
264 |
+
}
|
265 |
+
|
266 |
+
/* Markdown text styling */
|
267 |
+
.markdown-text {
|
268 |
+
line-height: 1.6;
|
269 |
+
font-size: 1.05rem;
|
270 |
+
color: #d0d0d0;
|
271 |
+
}
|
272 |
+
.markdown-text h1, .markdown-text h2, .markdown-text h3 {
|
273 |
+
color: #fff;
|
274 |
+
font-weight: 600;
|
275 |
+
margin-top: 1.5rem;
|
276 |
+
margin-bottom: 1rem;
|
277 |
+
}
|
278 |
+
.markdown-text p {
|
279 |
+
margin-bottom: 1rem;
|
280 |
+
}
|
281 |
"""
|
282 |
custom_css += dark_css
|
283 |
|
284 |
# --- Override Title ---
|
285 |
TITLE = """
|
286 |
+
<div class="hero-section">
|
287 |
+
<h1><i class="fa fa-trophy" style="color: gold;"></i> MLE-Dojo Benchmark Leaderboard</h1>
|
288 |
<h2>Improving LLM Agents for Machine Learning Engineering</h2>
|
289 |
</div>
|
290 |
"""
|
|
|
294 |
demo = gr.Blocks(css=custom_css, theme=gr.themes.Base())
|
295 |
|
296 |
with demo:
|
297 |
+
# Inject FontAwesome CSS and JS explicitly in the head
|
298 |
gr.HTML("""
|
299 |
+
<head>
|
300 |
+
<link rel="stylesheet"
|
301 |
+
href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css"
|
302 |
+
integrity="sha512-iecdLmaskl7CVkqkXNQ/ZH/XLlvWZOJyj7Yy7tcenmpD1ypASozpmT/E0iPtmFIB46ZmdtAc9eNBvH0H/ZpiBw=="
|
303 |
+
crossorigin="anonymous" referrerpolicy="no-referrer" />
|
304 |
+
<script src="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/js/all.min.js"
|
305 |
+
integrity="sha512-fD9DI5bZwQxOi7MhYWnnNPlvXdp/2Pj3XSTRrFs5FQa4mizyGLnJcN6tuvUS6LbmgN1ut+XGSABKvjN0H6Aoow=="
|
306 |
+
crossorigin="anonymous" referrerpolicy="no-referrer"></script>
|
307 |
+
</head>
|
308 |
+
""")
|
309 |
|
310 |
# Header & Intro
|
311 |
gr.HTML(TITLE)
|
312 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
313 |
|
314 |
+
# Tabs - Fix icons by using Font Awesome classes correctly
|
315 |
+
with gr.Tabs(elem_classes="tab-nav-buttons") as tabs:
|
316 |
+
with gr.TabItem("📋 Leaderboard"):
|
317 |
+
gr.Markdown("## Model Elo Rankings by Category", elem_classes="markdown-text")
|
318 |
category_selector = gr.Radio(
|
319 |
choices=CATEGORIES,
|
320 |
label="Select Category:",
|
|
|
338 |
outputs=leaderboard_df
|
339 |
)
|
340 |
|
341 |
+
with gr.TabItem("ℹ️ About"):
|
342 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
343 |
|
344 |
# Citation Accordion
|
|
|
353 |
|
354 |
if __name__ == "__main__":
|
355 |
print("Launching Gradio App in Dark Mode...")
|
356 |
+
demo.launch()
|