wangruisi1 commited on
Commit ·
fb265c6
1
Parent(s): 707db45
Update on table scheme
Browse files- app.py +130 -66
- src/display/css_html_js.py +30 -69
app.py
CHANGED
|
@@ -1,5 +1,6 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import pandas as pd
|
|
|
|
| 3 |
from collections import OrderedDict
|
| 4 |
|
| 5 |
from src.about import (
|
|
@@ -46,28 +47,34 @@ DEFAULT_GROUPS = [
|
|
| 46 |
ALWAYS_VISIBLE_COLS = ["Model", "Type"]
|
| 47 |
|
| 48 |
# ============================================================
|
| 49 |
-
# Column-to-color mapping
|
| 50 |
-
# Soft, harmonious palette: dark/light pairs for each group
|
| 51 |
# ============================================================
|
| 52 |
-
COLUMN_COLORS = {
|
| 53 |
-
# Overall (dark amber)
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
| 72 |
# ============================================================
|
| 73 |
# Static model scores data
|
|
@@ -86,6 +93,12 @@ MODEL_LINKS = {
|
|
| 86 |
"HunyuanVideo-I2V": "https://huggingface.co/tencent/HunyuanVideo-I2V",
|
| 87 |
}
|
| 88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
MODELS_DATA = [
|
| 90 |
{
|
| 91 |
"Model": "Human",
|
|
@@ -186,60 +199,105 @@ def build_full_dataframe():
|
|
| 186 |
# Round numeric columns to 3 decimal places for clean display
|
| 187 |
numeric_cols = df.select_dtypes(include="number").columns
|
| 188 |
df[numeric_cols] = df[numeric_cols].round(3)
|
|
|
|
|
|
|
| 189 |
return df
|
| 190 |
|
| 191 |
|
| 192 |
FULL_DF = build_full_dataframe()
|
| 193 |
|
| 194 |
|
| 195 |
-
def
|
| 196 |
-
"""
|
| 197 |
if not selected_groups:
|
| 198 |
-
selected_groups = ["Overall"]
|
| 199 |
|
| 200 |
cols = list(ALWAYS_VISIBLE_COLS)
|
| 201 |
for group_name, group_cols in COLUMN_GROUPS.items():
|
| 202 |
if group_name in selected_groups:
|
| 203 |
cols.extend(group_cols)
|
| 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 |
# ============================================================
|
|
@@ -260,15 +318,17 @@ with demo:
|
|
| 260 |
interactive=True,
|
| 261 |
)
|
| 262 |
|
| 263 |
-
|
| 264 |
-
value=
|
|
|
|
| 265 |
elem_id="leaderboard-table",
|
|
|
|
| 266 |
)
|
| 267 |
|
| 268 |
column_selector.change(
|
| 269 |
-
fn=
|
| 270 |
inputs=[column_selector],
|
| 271 |
-
outputs=[
|
| 272 |
)
|
| 273 |
|
| 274 |
with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=1):
|
|
@@ -287,4 +347,8 @@ with demo:
|
|
| 287 |
show_copy_button=True,
|
| 288 |
)
|
| 289 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 290 |
demo.queue(default_concurrency_limit=40).launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import pandas as pd
|
| 3 |
+
import json
|
| 4 |
from collections import OrderedDict
|
| 5 |
|
| 6 |
from src.about import (
|
|
|
|
| 47 |
ALWAYS_VISIBLE_COLS = ["Model", "Type"]
|
| 48 |
|
| 49 |
# ============================================================
|
| 50 |
+
# Column-to-color mapping (used by JS)
|
|
|
|
| 51 |
# ============================================================
|
| 52 |
+
COLUMN_COLORS = {
|
| 53 |
+
# Overall (dark amber)
|
| 54 |
+
"Overall": "rgba(232, 180, 58, 0.30)",
|
| 55 |
+
# Overall by Category (light amber)
|
| 56 |
+
"Abst.(All)": "rgba(242, 200, 90, 0.15)",
|
| 57 |
+
"Know.(All)": "rgba(242, 200, 90, 0.15)",
|
| 58 |
+
"Perc.(All)": "rgba(242, 200, 90, 0.15)",
|
| 59 |
+
"Spat.(All)": "rgba(242, 200, 90, 0.15)",
|
| 60 |
+
"Trans.(All)": "rgba(242, 200, 90, 0.15)",
|
| 61 |
+
# In-Domain Overall (dark green)
|
| 62 |
+
"Overall(In-Domain)": "rgba(82, 183, 120, 0.30)",
|
| 63 |
+
# In-Domain by Category (light green)
|
| 64 |
+
"Abst.(ID)": "rgba(110, 200, 145, 0.15)",
|
| 65 |
+
"Know.(ID)": "rgba(110, 200, 145, 0.15)",
|
| 66 |
+
"Perc.(ID)": "rgba(110, 200, 145, 0.15)",
|
| 67 |
+
"Spat.(ID)": "rgba(110, 200, 145, 0.15)",
|
| 68 |
+
"Trans.(ID)": "rgba(110, 200, 145, 0.15)",
|
| 69 |
+
# Out-of-Domain Overall (dark blue)
|
| 70 |
+
"Overall(Out-of-Domain)": "rgba(95, 150, 215, 0.30)",
|
| 71 |
+
# Out-of-Domain by Category (light blue)
|
| 72 |
+
"Abst.(OOD)": "rgba(125, 175, 228, 0.15)",
|
| 73 |
+
"Know.(OOD)": "rgba(125, 175, 228, 0.15)",
|
| 74 |
+
"Perc.(OOD)": "rgba(125, 175, 228, 0.15)",
|
| 75 |
+
"Spat.(OOD)": "rgba(125, 175, 228, 0.15)",
|
| 76 |
+
"Trans.(OOD)": "rgba(125, 175, 228, 0.15)",
|
| 77 |
+
}
|
| 78 |
|
| 79 |
# ============================================================
|
| 80 |
# Static model scores data
|
|
|
|
| 93 |
"HunyuanVideo-I2V": "https://huggingface.co/tencent/HunyuanVideo-I2V",
|
| 94 |
}
|
| 95 |
|
| 96 |
+
def make_model_link(model_name):
|
| 97 |
+
"""Create a clickable HTML link for a model if URL exists."""
|
| 98 |
+
if model_name in MODEL_LINKS:
|
| 99 |
+
return f'<a href="{MODEL_LINKS[model_name]}" target="_blank">{model_name}</a>'
|
| 100 |
+
return model_name
|
| 101 |
+
|
| 102 |
MODELS_DATA = [
|
| 103 |
{
|
| 104 |
"Model": "Human",
|
|
|
|
| 199 |
# Round numeric columns to 3 decimal places for clean display
|
| 200 |
numeric_cols = df.select_dtypes(include="number").columns
|
| 201 |
df[numeric_cols] = df[numeric_cols].round(3)
|
| 202 |
+
# Add clickable links to model names
|
| 203 |
+
df["Model"] = df["Model"].apply(make_model_link)
|
| 204 |
return df
|
| 205 |
|
| 206 |
|
| 207 |
FULL_DF = build_full_dataframe()
|
| 208 |
|
| 209 |
|
| 210 |
+
def get_filtered_df(selected_groups):
|
| 211 |
+
"""Filter DataFrame columns based on selected column groups."""
|
| 212 |
if not selected_groups:
|
| 213 |
+
selected_groups = ["Overall"] # Always show at least Overall
|
| 214 |
|
| 215 |
cols = list(ALWAYS_VISIBLE_COLS)
|
| 216 |
for group_name, group_cols in COLUMN_GROUPS.items():
|
| 217 |
if group_name in selected_groups:
|
| 218 |
cols.extend(group_cols)
|
| 219 |
|
| 220 |
+
return FULL_DF[cols]
|
| 221 |
+
|
| 222 |
+
|
| 223 |
+
# ============================================================
|
| 224 |
+
# Build the JS that colors columns by reading header text.
|
| 225 |
+
# Passed via Gradio's js= parameter on demo.load so it runs
|
| 226 |
+
# reliably after the page is fully rendered.
|
| 227 |
+
# ============================================================
|
| 228 |
+
COLOR_MAP_JSON = json.dumps(COLUMN_COLORS)
|
| 229 |
+
|
| 230 |
+
COLORING_JS = f"""
|
| 231 |
+
() => {{
|
| 232 |
+
const COLOR_MAP = {COLOR_MAP_JSON};
|
| 233 |
+
|
| 234 |
+
function colorColumns() {{
|
| 235 |
+
const container = document.querySelector('#leaderboard-table');
|
| 236 |
+
if (!container) return;
|
| 237 |
+
|
| 238 |
+
// Gradio Dataframe can use <table> or a virtual grid.
|
| 239 |
+
// Try standard <table> first.
|
| 240 |
+
const table = container.querySelector('table');
|
| 241 |
+
if (table) {{
|
| 242 |
+
const headers = table.querySelectorAll('thead th, thead td');
|
| 243 |
+
const headerTexts = [];
|
| 244 |
+
headers.forEach(th => headerTexts.push(th.textContent.trim()));
|
| 245 |
+
|
| 246 |
+
// Color header cells
|
| 247 |
+
headers.forEach((th, i) => {{
|
| 248 |
+
const color = COLOR_MAP[headerTexts[i]];
|
| 249 |
+
if (color) th.style.backgroundColor = color;
|
| 250 |
+
}});
|
| 251 |
+
|
| 252 |
+
// Color body cells
|
| 253 |
+
table.querySelectorAll('tbody tr').forEach(row => {{
|
| 254 |
+
const cells = row.querySelectorAll('td');
|
| 255 |
+
cells.forEach((td, i) => {{
|
| 256 |
+
const color = COLOR_MAP[headerTexts[i]];
|
| 257 |
+
if (color) td.style.backgroundColor = color;
|
| 258 |
+
}});
|
| 259 |
+
}});
|
| 260 |
+
return;
|
| 261 |
+
}}
|
| 262 |
+
|
| 263 |
+
// Fallback: Gradio virtual/svelte table (div-based grid)
|
| 264 |
+
const headerRow = container.querySelector('.header-row, .headers, [class*="header"]');
|
| 265 |
+
if (!headerRow) return;
|
| 266 |
+
const headerCells = headerRow.querySelectorAll('[class*="cell"], th, div');
|
| 267 |
+
const headerTexts = [];
|
| 268 |
+
headerCells.forEach(c => headerTexts.push(c.textContent.trim()));
|
| 269 |
+
|
| 270 |
+
headerCells.forEach((c, i) => {{
|
| 271 |
+
const color = COLOR_MAP[headerTexts[i]];
|
| 272 |
+
if (color) c.style.backgroundColor = color;
|
| 273 |
+
}});
|
| 274 |
+
|
| 275 |
+
const bodyRows = container.querySelectorAll('.body .row, tbody tr, [class*="row"]:not([class*="header"])');
|
| 276 |
+
bodyRows.forEach(row => {{
|
| 277 |
+
const cells = row.querySelectorAll('[class*="cell"], td, div');
|
| 278 |
+
cells.forEach((td, i) => {{
|
| 279 |
+
const color = COLOR_MAP[headerTexts[i]];
|
| 280 |
+
if (color) td.style.backgroundColor = color;
|
| 281 |
+
}});
|
| 282 |
+
}});
|
| 283 |
+
}}
|
| 284 |
+
|
| 285 |
+
// Run immediately, then with delays to catch late renders
|
| 286 |
+
colorColumns();
|
| 287 |
+
setTimeout(colorColumns, 300);
|
| 288 |
+
setTimeout(colorColumns, 800);
|
| 289 |
+
setTimeout(colorColumns, 1500);
|
| 290 |
+
|
| 291 |
+
// Also observe DOM changes to re-color when columns are toggled
|
| 292 |
+
const target = document.querySelector('#leaderboard-table');
|
| 293 |
+
if (target) {{
|
| 294 |
+
const obs = new MutationObserver(() => {{
|
| 295 |
+
setTimeout(colorColumns, 50);
|
| 296 |
+
}});
|
| 297 |
+
obs.observe(target, {{ childList: true, subtree: true }});
|
| 298 |
+
}}
|
| 299 |
+
}}
|
| 300 |
+
"""
|
| 301 |
|
| 302 |
|
| 303 |
# ============================================================
|
|
|
|
| 318 |
interactive=True,
|
| 319 |
)
|
| 320 |
|
| 321 |
+
leaderboard_table = gr.Dataframe(
|
| 322 |
+
value=get_filtered_df(DEFAULT_GROUPS),
|
| 323 |
+
interactive=False,
|
| 324 |
elem_id="leaderboard-table",
|
| 325 |
+
datatype=["html"] + ["str"] * 20,
|
| 326 |
)
|
| 327 |
|
| 328 |
column_selector.change(
|
| 329 |
+
fn=get_filtered_df,
|
| 330 |
inputs=[column_selector],
|
| 331 |
+
outputs=[leaderboard_table],
|
| 332 |
)
|
| 333 |
|
| 334 |
with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=1):
|
|
|
|
| 347 |
show_copy_button=True,
|
| 348 |
)
|
| 349 |
|
| 350 |
+
# Use Gradio's js= parameter on load — this is the official way
|
| 351 |
+
# to run JS after the page is fully rendered
|
| 352 |
+
demo.load(fn=None, inputs=None, outputs=None, js=COLORING_JS)
|
| 353 |
+
|
| 354 |
demo.queue(default_concurrency_limit=40).launch()
|
src/display/css_html_js.py
CHANGED
|
@@ -22,7 +22,11 @@ custom_css = """
|
|
| 22 |
}
|
| 23 |
|
| 24 |
#leaderboard-table {
|
| 25 |
-
margin-top: 15px
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
}
|
| 27 |
|
| 28 |
#search-bar-table-box > div:first-child {
|
|
@@ -34,6 +38,31 @@ custom_css = """
|
|
| 34 |
padding: 0px;
|
| 35 |
}
|
| 36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
.tab-buttons button {
|
| 38 |
font-size: 20px;
|
| 39 |
}
|
|
@@ -82,74 +111,6 @@ custom_css = """
|
|
| 82 |
#box-filter > .form{
|
| 83 |
border: 0
|
| 84 |
}
|
| 85 |
-
|
| 86 |
-
/* ============================================================
|
| 87 |
-
Styled HTML leaderboard table
|
| 88 |
-
============================================================ */
|
| 89 |
-
|
| 90 |
-
.leaderboard-scroll {
|
| 91 |
-
overflow-x: auto;
|
| 92 |
-
width: 100%;
|
| 93 |
-
}
|
| 94 |
-
|
| 95 |
-
.leaderboard-html-table {
|
| 96 |
-
width: 100%;
|
| 97 |
-
border-collapse: collapse;
|
| 98 |
-
font-family: 'Inter', -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
|
| 99 |
-
font-size: 14px;
|
| 100 |
-
line-height: 1.5;
|
| 101 |
-
}
|
| 102 |
-
|
| 103 |
-
.leaderboard-html-table thead th {
|
| 104 |
-
position: sticky;
|
| 105 |
-
top: 0;
|
| 106 |
-
z-index: 1;
|
| 107 |
-
padding: 10px 14px;
|
| 108 |
-
text-align: center;
|
| 109 |
-
font-weight: 600;
|
| 110 |
-
font-size: 13px;
|
| 111 |
-
border-bottom: 2px solid #d1d5db;
|
| 112 |
-
white-space: nowrap;
|
| 113 |
-
}
|
| 114 |
-
|
| 115 |
-
.leaderboard-html-table tbody td {
|
| 116 |
-
padding: 8px 14px;
|
| 117 |
-
text-align: center;
|
| 118 |
-
border-bottom: 1px solid #e5e7eb;
|
| 119 |
-
}
|
| 120 |
-
|
| 121 |
-
/* Model column — left-aligned, wider */
|
| 122 |
-
.leaderboard-html-table th:first-child,
|
| 123 |
-
.leaderboard-html-table td:first-child {
|
| 124 |
-
text-align: left;
|
| 125 |
-
min-width: 220px;
|
| 126 |
-
white-space: nowrap;
|
| 127 |
-
font-weight: 500;
|
| 128 |
-
}
|
| 129 |
-
|
| 130 |
-
/* Type column — left-aligned, wider */
|
| 131 |
-
.leaderboard-html-table th:nth-child(2),
|
| 132 |
-
.leaderboard-html-table td:nth-child(2) {
|
| 133 |
-
text-align: left;
|
| 134 |
-
min-width: 170px;
|
| 135 |
-
white-space: nowrap;
|
| 136 |
-
}
|
| 137 |
-
|
| 138 |
-
/* Row hover */
|
| 139 |
-
.leaderboard-html-table tbody tr:hover td {
|
| 140 |
-
filter: brightness(0.95);
|
| 141 |
-
}
|
| 142 |
-
|
| 143 |
-
/* Model links: underline + hover color change */
|
| 144 |
-
.leaderboard-html-table td a {
|
| 145 |
-
color: #2563eb;
|
| 146 |
-
text-decoration: underline;
|
| 147 |
-
text-underline-offset: 2px;
|
| 148 |
-
transition: color 0.2s ease;
|
| 149 |
-
}
|
| 150 |
-
.leaderboard-html-table td a:hover {
|
| 151 |
-
color: #d97706;
|
| 152 |
-
}
|
| 153 |
"""
|
| 154 |
|
| 155 |
get_window_url_params = """
|
|
|
|
| 22 |
}
|
| 23 |
|
| 24 |
#leaderboard-table {
|
| 25 |
+
margin-top: 15px
|
| 26 |
+
}
|
| 27 |
+
|
| 28 |
+
#leaderboard-table-lite {
|
| 29 |
+
margin-top: 15px
|
| 30 |
}
|
| 31 |
|
| 32 |
#search-bar-table-box > div:first-child {
|
|
|
|
| 38 |
padding: 0px;
|
| 39 |
}
|
| 40 |
|
| 41 |
+
/* Model column — wider to avoid truncation */
|
| 42 |
+
#leaderboard-table td:nth-child(1),
|
| 43 |
+
#leaderboard-table th:nth-child(1) {
|
| 44 |
+
min-width: 220px !important;
|
| 45 |
+
white-space: nowrap;
|
| 46 |
+
}
|
| 47 |
+
|
| 48 |
+
/* Type column — wider to avoid truncation */
|
| 49 |
+
#leaderboard-table td:nth-child(2),
|
| 50 |
+
#leaderboard-table th:nth-child(2) {
|
| 51 |
+
min-width: 170px !important;
|
| 52 |
+
white-space: nowrap;
|
| 53 |
+
}
|
| 54 |
+
|
| 55 |
+
/* Model links: underline + hover color change */
|
| 56 |
+
#leaderboard-table td a {
|
| 57 |
+
color: #2563eb;
|
| 58 |
+
text-decoration: underline;
|
| 59 |
+
text-underline-offset: 2px;
|
| 60 |
+
transition: color 0.2s ease;
|
| 61 |
+
}
|
| 62 |
+
#leaderboard-table td a:hover {
|
| 63 |
+
color: #d97706;
|
| 64 |
+
}
|
| 65 |
+
|
| 66 |
.tab-buttons button {
|
| 67 |
font-size: 20px;
|
| 68 |
}
|
|
|
|
| 111 |
#box-filter > .form{
|
| 112 |
border: 0
|
| 113 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
"""
|
| 115 |
|
| 116 |
get_window_url_params = """
|