Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -7,6 +7,8 @@ os.environ["GRADIO_LANGUAGE"] = "en"
|
|
| 7 |
|
| 8 |
RESULT_DIR = os.environ.get("MOECAP_RESULT_DIR")
|
| 9 |
if not RESULT_DIR:
|
|
|
|
|
|
|
| 10 |
raise RuntimeError(
|
| 11 |
"MOECAP_RESULT_DIR is not set. Please set MOECAP_RESULT_DIR (HF Repo ID) before running app.py"
|
| 12 |
)
|
|
@@ -48,12 +50,16 @@ def normalize_cost(val, max_tick, baseline=20):
|
|
| 48 |
def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
| 49 |
"""Generate a CAP radar plot from selected rows."""
|
| 50 |
|
| 51 |
-
#
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
if not selected_rows_data or len(selected_rows_data) == 0:
|
| 58 |
fig = go.Figure()
|
| 59 |
fig.add_annotation(
|
|
@@ -65,12 +71,9 @@ def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
|
| 65 |
yanchor='middle'
|
| 66 |
)
|
| 67 |
fig.update_layout(
|
| 68 |
-
height=600, # Reduced slightly to fit screens better
|
| 69 |
xaxis=dict(visible=False),
|
| 70 |
yaxis=dict(visible=False),
|
| 71 |
-
|
| 72 |
-
plot_bgcolor='white',
|
| 73 |
-
margin=common_margin # Use balanced margins
|
| 74 |
)
|
| 75 |
return fig
|
| 76 |
|
|
@@ -85,12 +88,9 @@ def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
|
| 85 |
yanchor='middle'
|
| 86 |
)
|
| 87 |
fig.update_layout(
|
| 88 |
-
height=600,
|
| 89 |
xaxis=dict(visible=False),
|
| 90 |
yaxis=dict(visible=False),
|
| 91 |
-
|
| 92 |
-
plot_bgcolor='white',
|
| 93 |
-
margin=common_margin # Use balanced margins
|
| 94 |
)
|
| 95 |
return fig
|
| 96 |
|
|
@@ -107,12 +107,9 @@ def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
|
| 107 |
yanchor='middle'
|
| 108 |
)
|
| 109 |
fig.update_layout(
|
| 110 |
-
height=600,
|
| 111 |
xaxis=dict(visible=False),
|
| 112 |
yaxis=dict(visible=False),
|
| 113 |
-
|
| 114 |
-
plot_bgcolor='white',
|
| 115 |
-
margin=common_margin # Use balanced margins
|
| 116 |
)
|
| 117 |
return fig
|
| 118 |
|
|
@@ -203,7 +200,7 @@ def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
|
| 203 |
title=dict(
|
| 204 |
text=f"CAP Radar Plot: {dataset_name}",
|
| 205 |
x=0.5,
|
| 206 |
-
xanchor='center',
|
| 207 |
font=dict(size=20)
|
| 208 |
),
|
| 209 |
polar=dict(
|
|
@@ -214,25 +211,19 @@ def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
|
| 214 |
),
|
| 215 |
angularaxis=dict(
|
| 216 |
tickfont=dict(size=14),
|
| 217 |
-
rotation=90, #
|
| 218 |
direction='clockwise'
|
| 219 |
),
|
| 220 |
),
|
| 221 |
legend=dict(
|
| 222 |
orientation='h',
|
| 223 |
yanchor='bottom',
|
| 224 |
-
y=-0.15,
|
| 225 |
xanchor='center',
|
| 226 |
x=0.5,
|
| 227 |
font=dict(size=13)
|
| 228 |
),
|
| 229 |
-
|
| 230 |
-
margin=common_margin,
|
| 231 |
-
height=700,
|
| 232 |
-
# Removed fixed 'width' to allow Gradio to resize it responsively
|
| 233 |
-
autosize=True,
|
| 234 |
-
paper_bgcolor='white',
|
| 235 |
-
plot_bgcolor='white'
|
| 236 |
)
|
| 237 |
|
| 238 |
return fig
|
|
@@ -289,83 +280,11 @@ def json_to_row(path: str, metrics: dict) -> dict:
|
|
| 289 |
"Decoding<br>S-MFU(%)": pct(metrics.get("decoding_smfu")),
|
| 290 |
"TTFT(s)": f2(metrics.get("ttft")),
|
| 291 |
"TPOT(s)": f2(metrics.get("tpot")),
|
| 292 |
-
"Batch size": batch_size,
|
| 293 |
}
|
| 294 |
return row
|
| 295 |
|
| 296 |
|
| 297 |
-
def build_leaderboard_from_files(files: List[gr.File], prev_rows: list | None = None):
|
| 298 |
-
if prev_rows is None:
|
| 299 |
-
prev_rows = []
|
| 300 |
-
|
| 301 |
-
if not files and prev_rows:
|
| 302 |
-
df = pd.DataFrame(prev_rows)
|
| 303 |
-
raw_models = set()
|
| 304 |
-
for cell in df["Model"].tolist():
|
| 305 |
-
if isinstance(cell, str) and "href" in cell:
|
| 306 |
-
try:
|
| 307 |
-
name = cell.split(">", 1)[1].split("<", 1)[0]
|
| 308 |
-
except Exception:
|
| 309 |
-
name = cell
|
| 310 |
-
else:
|
| 311 |
-
name = cell
|
| 312 |
-
raw_models.add(name)
|
| 313 |
-
links = []
|
| 314 |
-
for name in sorted(raw_models):
|
| 315 |
-
if isinstance(name, str) and "/" in name:
|
| 316 |
-
hf_url = f"https://huggingface.co/{name}"
|
| 317 |
-
links.append(f"[{name}]({hf_url})")
|
| 318 |
-
else:
|
| 319 |
-
links.append(str(name))
|
| 320 |
-
models_str = ", ".join(links)
|
| 321 |
-
summary_md = f"**Loaded {len(prev_rows)} result files.** \n**Models:** {models_str}"
|
| 322 |
-
table_html = f'<div class="table-container">{df.to_html(escape=False, index=False, classes="metrics-table")}</div>'
|
| 323 |
-
return summary_md, table_html, prev_rows
|
| 324 |
-
|
| 325 |
-
new_rows = []
|
| 326 |
-
if files:
|
| 327 |
-
for f in files:
|
| 328 |
-
path = f.name
|
| 329 |
-
try:
|
| 330 |
-
with open(path, "r", encoding="utf-8") as fp:
|
| 331 |
-
metrics = json.load(fp)
|
| 332 |
-
new_rows.append(json_to_row(path, metrics))
|
| 333 |
-
except Exception:
|
| 334 |
-
continue
|
| 335 |
-
|
| 336 |
-
all_rows = prev_rows + new_rows
|
| 337 |
-
|
| 338 |
-
if not all_rows:
|
| 339 |
-
empty_html = "<p>No files loaded.</p>"
|
| 340 |
-
return "No files uploaded.", empty_html, []
|
| 341 |
-
|
| 342 |
-
df = pd.DataFrame(all_rows)
|
| 343 |
-
|
| 344 |
-
raw_models = set()
|
| 345 |
-
for cell in df["Model"].tolist():
|
| 346 |
-
if isinstance(cell, str) and "href" in cell:
|
| 347 |
-
try:
|
| 348 |
-
name = cell.split(">", 1)[1].split("<", 1)[0]
|
| 349 |
-
except Exception:
|
| 350 |
-
name = cell
|
| 351 |
-
else:
|
| 352 |
-
name = cell
|
| 353 |
-
raw_models.add(name)
|
| 354 |
-
links = []
|
| 355 |
-
for name in sorted(raw_models):
|
| 356 |
-
if isinstance(name, str) and "/" in name:
|
| 357 |
-
hf_url = f"https://huggingface.co/{name}"
|
| 358 |
-
links.append(f"[{name}]({hf_url})")
|
| 359 |
-
else:
|
| 360 |
-
links.append(str(name))
|
| 361 |
-
models_str = ", ".join(links)
|
| 362 |
-
summary_md = f"**Loaded {len(all_rows)} result files.** \n**Models:** {models_str}"
|
| 363 |
-
|
| 364 |
-
table_html = f'<div class="table-container">{df.to_html(escape=False, index=False, classes="metrics-table")}</div>'
|
| 365 |
-
|
| 366 |
-
return summary_md, table_html, all_rows
|
| 367 |
-
|
| 368 |
-
|
| 369 |
def load_from_dir(
|
| 370 |
dir_path: str,
|
| 371 |
selected_tasks: List[str] | None = None,
|
|
@@ -388,7 +307,7 @@ def load_from_dir(
|
|
| 388 |
)
|
| 389 |
except Exception as e:
|
| 390 |
empty_html = "<p>No files loaded or Dataset not found.</p>"
|
| 391 |
-
return empty_html
|
| 392 |
|
| 393 |
rows = []
|
| 394 |
for i, example in enumerate(ds):
|
|
@@ -400,35 +319,29 @@ def load_from_dir(
|
|
| 400 |
|
| 401 |
if not rows:
|
| 402 |
empty_html = "<p>No records found.</p>"
|
| 403 |
-
return empty_html
|
| 404 |
|
| 405 |
df = pd.DataFrame(rows)
|
| 406 |
|
| 407 |
-
#
|
| 408 |
if selected_tasks is not None:
|
| 409 |
lower_selected = [x.lower() for x in selected_tasks]
|
| 410 |
df = df[df["Dataset"].astype(str).str.lower().isin(lower_selected)]
|
| 411 |
|
| 412 |
-
|
| 413 |
-
# Inference framework filter (Method)
|
| 414 |
if selected_frameworks is not None:
|
| 415 |
lower_selected = [str(x).lower() for x in selected_frameworks]
|
| 416 |
df = df[df["Method"].astype(str).str.lower().isin(lower_selected)]
|
| 417 |
|
| 418 |
-
# Model type filter
|
| 419 |
if selected_model_types is not None:
|
| 420 |
lower_selected = [str(x).lower() for x in selected_model_types]
|
| 421 |
df = df[df["Model type"].astype(str).str.lower().isin(lower_selected)]
|
| 422 |
|
| 423 |
-
# Precision filter
|
| 424 |
if selected_precisions is not None:
|
| 425 |
lower_selected = [str(x).lower() for x in selected_precisions]
|
| 426 |
df = df[df["Precision"].astype(str).str.lower().isin(lower_selected)]
|
| 427 |
|
| 428 |
-
# Search keyword filter - search across all columns
|
| 429 |
if search_keyword and search_keyword.strip():
|
| 430 |
keyword_lower = search_keyword.strip().lower()
|
| 431 |
-
# Create a mask that checks if the keyword appears in any column
|
| 432 |
mask = df.astype(str).apply(lambda row: row.str.lower().str.contains(keyword_lower).any(), axis=1)
|
| 433 |
df = df[mask]
|
| 434 |
|
|
@@ -437,29 +350,8 @@ def load_from_dir(
|
|
| 437 |
return empty_html, []
|
| 438 |
|
| 439 |
df = df.fillna("-")
|
| 440 |
-
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
for cell in df["Model"].tolist():
|
| 444 |
-
if isinstance(cell, str) and "href" in cell:
|
| 445 |
-
try:
|
| 446 |
-
name = cell.split(">", 1)[1].split("<", 1)[0]
|
| 447 |
-
except Exception:
|
| 448 |
-
name = cell
|
| 449 |
-
else:
|
| 450 |
-
name = cell
|
| 451 |
-
raw_models.add(name)
|
| 452 |
-
|
| 453 |
-
links = []
|
| 454 |
-
for name in sorted(raw_models):
|
| 455 |
-
if isinstance(name, str) and "/" in name:
|
| 456 |
-
hf_url = f"https://huggingface.co/{name}"
|
| 457 |
-
links.append(f"[{name}]({hf_url})")
|
| 458 |
-
else:
|
| 459 |
-
links.append(str(name))
|
| 460 |
-
models_str = ", ".join(links)
|
| 461 |
-
|
| 462 |
-
# Insert row number column at the beginning for easy reference
|
| 463 |
df.insert(0, 'Row #', range(len(df)))
|
| 464 |
|
| 465 |
# Create HTML table
|
|
@@ -488,16 +380,6 @@ def auto_refresh_from_dir(
|
|
| 488 |
)
|
| 489 |
|
| 490 |
|
| 491 |
-
def update_radar_plot(df_data: list, selected_indices: list):
|
| 492 |
-
"""Update radar plot based on selected row indices."""
|
| 493 |
-
if not selected_indices or not df_data:
|
| 494 |
-
return generate_radar_plot([])
|
| 495 |
-
|
| 496 |
-
# Get selected rows (limit to 3)
|
| 497 |
-
selected_rows = [df_data[i] for i in selected_indices[:3] if i < len(df_data)]
|
| 498 |
-
return generate_radar_plot(selected_rows)
|
| 499 |
-
|
| 500 |
-
|
| 501 |
def parse_and_generate_plot(df_data: list, indices_str: str):
|
| 502 |
"""Parse comma-separated indices and generate radar plot."""
|
| 503 |
if not indices_str or not indices_str.strip():
|
|
@@ -515,273 +397,79 @@ def parse_and_generate_plot(df_data: list, indices_str: str):
|
|
| 515 |
return generate_radar_plot([])
|
| 516 |
|
| 517 |
|
| 518 |
-
def on_table_select(df, evt: gr.SelectData):
|
| 519 |
-
"""Handle table row selection."""
|
| 520 |
-
return evt.index
|
| 521 |
-
|
| 522 |
-
|
| 523 |
# Gradio UI
|
| 524 |
|
| 525 |
def build_app() -> gr.Blocks:
|
| 526 |
row_css = """
|
| 527 |
-
|
| 528 |
-
body {
|
| 529 |
-
background-color: #f5f7fa !important;
|
| 530 |
-
}
|
| 531 |
|
| 532 |
/* Row number column styling */
|
| 533 |
-
.metrics-table th:first-child,
|
| 534 |
-
|
| 535 |
-
|
| 536 |
-
text-align: center !important;
|
| 537 |
-
padding: 8px !important;
|
| 538 |
-
font-weight: 600 !important;
|
| 539 |
background-color: #f0f0f0 !important;
|
| 540 |
}
|
| 541 |
-
/* The outer Group container */
|
| 542 |
.search-box {
|
| 543 |
-
background: white !important;
|
| 544 |
-
|
| 545 |
-
|
| 546 |
-
border: 2px solid #e1e4e8 !important;
|
| 547 |
-
box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
| 548 |
-
margin-bottom: 16px;
|
| 549 |
}
|
| 550 |
-
|
| 551 |
-
.search-box
|
| 552 |
-
background: transparent !important;
|
| 553 |
-
border: none !important;
|
| 554 |
-
padding: 0 !important;
|
| 555 |
-
}
|
| 556 |
-
/* Style the Label Text (🔍 Search) */
|
| 557 |
-
.search-box label span {
|
| 558 |
-
color: #24292e !important;
|
| 559 |
-
font-weight: 600;
|
| 560 |
-
font-size: 14px;
|
| 561 |
-
margin-bottom: 8px;
|
| 562 |
-
background: transparent !important;
|
| 563 |
-
}
|
| 564 |
-
/* Style the actual Input Field */
|
| 565 |
.search-box input.scroll-hide {
|
| 566 |
-
background-color: white !important;
|
| 567 |
-
|
| 568 |
-
|
| 569 |
-
border-radius: 4px !important;
|
| 570 |
-
padding: 10px !important;
|
| 571 |
-
box-shadow: none !important;
|
| 572 |
-
}
|
| 573 |
-
/* Fix focus state */
|
| 574 |
-
.search-box input.scroll-hide:focus {
|
| 575 |
-
border-color: #0366d6 !important;
|
| 576 |
-
ring: 0 !important;
|
| 577 |
-
outline: none !important;
|
| 578 |
-
}
|
| 579 |
-
|
| 580 |
-
.gradio-container {
|
| 581 |
-
max-width: 100% !important;
|
| 582 |
-
padding: 20px !important;
|
| 583 |
-
background-color: #f5f7fa !important;
|
| 584 |
-
}
|
| 585 |
-
|
| 586 |
-
/* Override all dark backgrounds */
|
| 587 |
-
.gradio-container .block,
|
| 588 |
-
.gradio-container .form,
|
| 589 |
-
.gradio-container fieldset,
|
| 590 |
-
.gradio-container .input-block,
|
| 591 |
-
.gradio-container .wrap,
|
| 592 |
-
.gradio-container .gr-box,
|
| 593 |
-
.gradio-container .gr-form,
|
| 594 |
-
.gradio-container .gr-input {
|
| 595 |
-
background-color: white !important;
|
| 596 |
-
border-color: #e1e4e8 !important;
|
| 597 |
-
}
|
| 598 |
-
|
| 599 |
-
.gradio-container label {
|
| 600 |
-
background-color: transparent !important;
|
| 601 |
-
color: #24292e !important;
|
| 602 |
-
}
|
| 603 |
-
|
| 604 |
-
/* Remove any potential dark wrappers */
|
| 605 |
-
.gradio-container > div,
|
| 606 |
-
.gradio-container .container {
|
| 607 |
-
background-color: transparent !important;
|
| 608 |
}
|
|
|
|
| 609 |
|
| 610 |
-
|
| 611 |
-
.gradio-container,
|
| 612 |
-
|
| 613 |
-
.gradio-container p,
|
| 614 |
-
.gradio-container span,
|
| 615 |
-
.gradio-container div {
|
| 616 |
-
color: #24292e !important;
|
| 617 |
}
|
|
|
|
| 618 |
|
| 619 |
/* Table styling */
|
| 620 |
-
.gradio-container table.metrics-table th,
|
| 621 |
-
|
| 622 |
-
|
| 623 |
-
border: 1.5px solid #e1e4e8;
|
| 624 |
-
white-space: nowrap;
|
| 625 |
-
font-size: 13px;
|
| 626 |
-
text-align: left;
|
| 627 |
-
color: #24292e !important;
|
| 628 |
}
|
| 629 |
-
|
| 630 |
.gradio-container table.metrics-table th {
|
| 631 |
background: linear-gradient(to bottom, #fafbfc, #f6f8fa);
|
| 632 |
-
font-weight: 600;
|
| 633 |
-
color: #24292e !important;
|
| 634 |
-
position: sticky;
|
| 635 |
-
top: 0;
|
| 636 |
-
z-index: 10;
|
| 637 |
border-bottom: 2px solid #d1d5da;
|
| 638 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 639 |
|
| 640 |
-
|
| 641 |
-
|
| 642 |
-
}
|
| 643 |
-
|
| 644 |
-
.gradio-container table.metrics-table tbody tr:hover {
|
| 645 |
-
background-color: #e1e4e8;
|
| 646 |
-
}
|
| 647 |
-
|
| 648 |
-
.gradio-container table.metrics-table {
|
| 649 |
-
border-collapse: collapse;
|
| 650 |
-
width: 100%;
|
| 651 |
-
background: white;
|
| 652 |
-
}
|
| 653 |
-
|
| 654 |
-
.gradio-container table.metrics-table a {
|
| 655 |
-
color: #0366d6 !important;
|
| 656 |
-
text-decoration: none;
|
| 657 |
-
}
|
| 658 |
-
|
| 659 |
-
.gradio-container table.metrics-table a:hover {
|
| 660 |
-
color: #0366d6 !important;
|
| 661 |
-
text-decoration: underline;
|
| 662 |
-
}
|
| 663 |
-
|
| 664 |
-
.gradio-container .plot-container {
|
| 665 |
-
width: 100% !important;
|
| 666 |
-
max-width: 1200px !important;
|
| 667 |
-
}
|
| 668 |
|
| 669 |
-
/* Scrollable table container */
|
| 670 |
.table-container {
|
| 671 |
-
overflow-x: auto;
|
| 672 |
-
|
| 673 |
-
|
| 674 |
-
border: 2px solid #e1e4e8;
|
| 675 |
-
border-radius: 6px;
|
| 676 |
-
background: white;
|
| 677 |
-
box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
| 678 |
}
|
| 679 |
-
|
| 680 |
-
/* Filter section styling */
|
| 681 |
.filter-section {
|
| 682 |
-
background: white !important;
|
| 683 |
-
|
| 684 |
-
border-radius: 6px;
|
| 685 |
-
border: 2px solid #e1e4e8 !important;
|
| 686 |
-
box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
| 687 |
-
}
|
| 688 |
-
|
| 689 |
-
.filter-section * {
|
| 690 |
-
color: #24292e !important;
|
| 691 |
-
}
|
| 692 |
-
|
| 693 |
-
.filter-section .wrap,
|
| 694 |
-
.filter-section .block,
|
| 695 |
-
.filter-section .container,
|
| 696 |
-
.filter-section .group,
|
| 697 |
-
.filter-section > div,
|
| 698 |
-
.filter-section > div > div {
|
| 699 |
-
background: transparent !important;
|
| 700 |
-
}
|
| 701 |
-
|
| 702 |
-
.filter-section .wrap {
|
| 703 |
-
padding: 20px !important;
|
| 704 |
-
}
|
| 705 |
-
|
| 706 |
-
.filter-section label {
|
| 707 |
-
background: transparent !important;
|
| 708 |
-
color: #24292e !important;
|
| 709 |
}
|
| 710 |
-
|
| 711 |
-
.filter-section
|
| 712 |
-
background: transparent !important;
|
| 713 |
-
border-color: #e1e4e8 !important;
|
| 714 |
-
}
|
| 715 |
-
|
| 716 |
-
/* Accordion styling */
|
| 717 |
.gradio-container .accordion {
|
| 718 |
-
background: white !important;
|
| 719 |
-
border:
|
| 720 |
-
|
| 721 |
-
|
| 722 |
-
}
|
| 723 |
-
|
| 724 |
-
.gradio-container
|
| 725 |
-
|
| 726 |
-
}
|
| 727 |
-
|
| 728 |
-
.gradio-container .accordion label {
|
| 729 |
-
background: transparent !important;
|
| 730 |
-
color: #24292e !important;
|
| 731 |
-
}
|
| 732 |
-
|
| 733 |
-
.gradio-container .accordion button {
|
| 734 |
-
background: transparent !important;
|
| 735 |
-
color: #24292e !important;
|
| 736 |
-
}
|
| 737 |
-
|
| 738 |
-
/* Info section */
|
| 739 |
-
.info-section {
|
| 740 |
-
padding: 16px;
|
| 741 |
-
background: white !important;
|
| 742 |
-
}
|
| 743 |
-
|
| 744 |
-
/* Make text in info section dark */
|
| 745 |
-
.info-section p,
|
| 746 |
-
.info-section li,
|
| 747 |
-
.info-section ul,
|
| 748 |
-
.info-section h3,
|
| 749 |
-
.info-section strong,
|
| 750 |
-
.info-section * {
|
| 751 |
-
color: #24292e !important;
|
| 752 |
-
}
|
| 753 |
-
|
| 754 |
-
.info-section a {
|
| 755 |
-
color: #0366d6 !important;
|
| 756 |
-
}
|
| 757 |
-
|
| 758 |
-
/* Override any dark backgrounds in groups and accordions */
|
| 759 |
-
.gradio-container .group,
|
| 760 |
-
.gradio-container .accordion,
|
| 761 |
-
.gradio-container .panel {
|
| 762 |
-
background-color: white !important;
|
| 763 |
-
}
|
| 764 |
-
|
| 765 |
-
/* Heading styling */
|
| 766 |
-
.gradio-container h1 {
|
| 767 |
-
color: #24292e !important;
|
| 768 |
-
font-weight: 700;
|
| 769 |
-
margin-bottom: 24px;
|
| 770 |
-
}
|
| 771 |
-
|
| 772 |
-
.gradio-container h3 {
|
| 773 |
-
color: #24292e !important;
|
| 774 |
-
font-weight: 600;
|
| 775 |
-
margin-bottom: 16px;
|
| 776 |
-
}
|
| 777 |
-
|
| 778 |
-
/* Checkbox styling */
|
| 779 |
-
.gradio-container input[type="checkbox"] {
|
| 780 |
-
accent-color: #0366d6 !important;
|
| 781 |
-
}
|
| 782 |
"""
|
| 783 |
|
| 784 |
-
# Use Gradio's default (light) theme explicitly
|
| 785 |
with gr.Blocks(title="MoE-CAP Dashboard", css=row_css, theme=gr.themes.Default()) as demo:
|
| 786 |
gr.Markdown("# MoE-CAP Dashboard")
|
| 787 |
|
|
@@ -800,7 +488,6 @@ def build_app() -> gr.Blocks:
|
|
| 800 |
|
| 801 |
dir_path = gr.State(RESULT_DIR)
|
| 802 |
|
| 803 |
-
# 1) Tasks filter
|
| 804 |
task_filter = gr.CheckboxGroup(
|
| 805 |
label="📊 Tasks",
|
| 806 |
choices=[
|
|
@@ -813,21 +500,18 @@ def build_app() -> gr.Blocks:
|
|
| 813 |
value=["gsm8k", "longbench", "mmlu", "numinamath", "ruler"]
|
| 814 |
)
|
| 815 |
|
| 816 |
-
# 2) Inference frameworks filter
|
| 817 |
framework_filter = gr.CheckboxGroup(
|
| 818 |
label="⚙️ Inference Frameworks",
|
| 819 |
choices=["sglang", "vllm"],
|
| 820 |
value=["sglang", "vllm"],
|
| 821 |
)
|
| 822 |
|
| 823 |
-
# 3) Model types filter
|
| 824 |
model_type_filter = gr.CheckboxGroup(
|
| 825 |
label="🤖 Model Types",
|
| 826 |
choices=["instruct", "thinking"],
|
| 827 |
value=["instruct", "thinking"],
|
| 828 |
)
|
| 829 |
|
| 830 |
-
# 4) Precision filter
|
| 831 |
precision_filter = gr.CheckboxGroup(
|
| 832 |
label="🎯 Precision",
|
| 833 |
choices=["bfloat16", "fp8"],
|
|
@@ -874,13 +558,12 @@ def build_app() -> gr.Blocks:
|
|
| 874 |
)
|
| 875 |
generate_btn = gr.Button("🎯 Generate", variant="primary", scale=1, size="lg")
|
| 876 |
|
| 877 |
-
|
| 878 |
-
|
| 879 |
-
|
| 880 |
-
|
| 881 |
-
|
| 882 |
-
|
| 883 |
-
pass
|
| 884 |
|
| 885 |
df_data_state = gr.State([])
|
| 886 |
|
|
|
|
| 7 |
|
| 8 |
RESULT_DIR = os.environ.get("MOECAP_RESULT_DIR")
|
| 9 |
if not RESULT_DIR:
|
| 10 |
+
# For testing purposes, you can uncomment the line below to set a dummy dir or keep the raise
|
| 11 |
+
# RESULT_DIR = "generic_result_dir"
|
| 12 |
raise RuntimeError(
|
| 13 |
"MOECAP_RESULT_DIR is not set. Please set MOECAP_RESULT_DIR (HF Repo ID) before running app.py"
|
| 14 |
)
|
|
|
|
| 50 |
def generate_radar_plot(selected_rows_data: List[dict]) -> go.Figure:
|
| 51 |
"""Generate a CAP radar plot from selected rows."""
|
| 52 |
|
| 53 |
+
# Standard layout settings for consistent sizing
|
| 54 |
+
layout_settings = dict(
|
| 55 |
+
height=750, # Taller height
|
| 56 |
+
autosize=True, # Auto width
|
| 57 |
+
margin=dict(t=80, b=100, l=80, r=80), # Balanced margins
|
| 58 |
+
paper_bgcolor='white',
|
| 59 |
+
plot_bgcolor='white',
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
# Validation: max 3 rows
|
| 63 |
if not selected_rows_data or len(selected_rows_data) == 0:
|
| 64 |
fig = go.Figure()
|
| 65 |
fig.add_annotation(
|
|
|
|
| 71 |
yanchor='middle'
|
| 72 |
)
|
| 73 |
fig.update_layout(
|
|
|
|
| 74 |
xaxis=dict(visible=False),
|
| 75 |
yaxis=dict(visible=False),
|
| 76 |
+
**layout_settings
|
|
|
|
|
|
|
| 77 |
)
|
| 78 |
return fig
|
| 79 |
|
|
|
|
| 88 |
yanchor='middle'
|
| 89 |
)
|
| 90 |
fig.update_layout(
|
|
|
|
| 91 |
xaxis=dict(visible=False),
|
| 92 |
yaxis=dict(visible=False),
|
| 93 |
+
**layout_settings
|
|
|
|
|
|
|
| 94 |
)
|
| 95 |
return fig
|
| 96 |
|
|
|
|
| 107 |
yanchor='middle'
|
| 108 |
)
|
| 109 |
fig.update_layout(
|
|
|
|
| 110 |
xaxis=dict(visible=False),
|
| 111 |
yaxis=dict(visible=False),
|
| 112 |
+
**layout_settings
|
|
|
|
|
|
|
| 113 |
)
|
| 114 |
return fig
|
| 115 |
|
|
|
|
| 200 |
title=dict(
|
| 201 |
text=f"CAP Radar Plot: {dataset_name}",
|
| 202 |
x=0.5,
|
| 203 |
+
xanchor='center',
|
| 204 |
font=dict(size=20)
|
| 205 |
),
|
| 206 |
polar=dict(
|
|
|
|
| 211 |
),
|
| 212 |
angularaxis=dict(
|
| 213 |
tickfont=dict(size=14),
|
| 214 |
+
rotation=90, # Rotate so top is 12 o'clock
|
| 215 |
direction='clockwise'
|
| 216 |
),
|
| 217 |
),
|
| 218 |
legend=dict(
|
| 219 |
orientation='h',
|
| 220 |
yanchor='bottom',
|
| 221 |
+
y=-0.15,
|
| 222 |
xanchor='center',
|
| 223 |
x=0.5,
|
| 224 |
font=dict(size=13)
|
| 225 |
),
|
| 226 |
+
**layout_settings
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
)
|
| 228 |
|
| 229 |
return fig
|
|
|
|
| 280 |
"Decoding<br>S-MFU(%)": pct(metrics.get("decoding_smfu")),
|
| 281 |
"TTFT(s)": f2(metrics.get("ttft")),
|
| 282 |
"TPOT(s)": f2(metrics.get("tpot")),
|
| 283 |
+
"Batch size": batch_size,
|
| 284 |
}
|
| 285 |
return row
|
| 286 |
|
| 287 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 288 |
def load_from_dir(
|
| 289 |
dir_path: str,
|
| 290 |
selected_tasks: List[str] | None = None,
|
|
|
|
| 307 |
)
|
| 308 |
except Exception as e:
|
| 309 |
empty_html = "<p>No files loaded or Dataset not found.</p>"
|
| 310 |
+
return empty_html, []
|
| 311 |
|
| 312 |
rows = []
|
| 313 |
for i, example in enumerate(ds):
|
|
|
|
| 319 |
|
| 320 |
if not rows:
|
| 321 |
empty_html = "<p>No records found.</p>"
|
| 322 |
+
return empty_html, []
|
| 323 |
|
| 324 |
df = pd.DataFrame(rows)
|
| 325 |
|
| 326 |
+
# Filters
|
| 327 |
if selected_tasks is not None:
|
| 328 |
lower_selected = [x.lower() for x in selected_tasks]
|
| 329 |
df = df[df["Dataset"].astype(str).str.lower().isin(lower_selected)]
|
| 330 |
|
|
|
|
|
|
|
| 331 |
if selected_frameworks is not None:
|
| 332 |
lower_selected = [str(x).lower() for x in selected_frameworks]
|
| 333 |
df = df[df["Method"].astype(str).str.lower().isin(lower_selected)]
|
| 334 |
|
|
|
|
| 335 |
if selected_model_types is not None:
|
| 336 |
lower_selected = [str(x).lower() for x in selected_model_types]
|
| 337 |
df = df[df["Model type"].astype(str).str.lower().isin(lower_selected)]
|
| 338 |
|
|
|
|
| 339 |
if selected_precisions is not None:
|
| 340 |
lower_selected = [str(x).lower() for x in selected_precisions]
|
| 341 |
df = df[df["Precision"].astype(str).str.lower().isin(lower_selected)]
|
| 342 |
|
|
|
|
| 343 |
if search_keyword and search_keyword.strip():
|
| 344 |
keyword_lower = search_keyword.strip().lower()
|
|
|
|
| 345 |
mask = df.astype(str).apply(lambda row: row.str.lower().str.contains(keyword_lower).any(), axis=1)
|
| 346 |
df = df[mask]
|
| 347 |
|
|
|
|
| 350 |
return empty_html, []
|
| 351 |
|
| 352 |
df = df.fillna("-")
|
| 353 |
+
|
| 354 |
+
# Insert row number column at the beginning
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 355 |
df.insert(0, 'Row #', range(len(df)))
|
| 356 |
|
| 357 |
# Create HTML table
|
|
|
|
| 380 |
)
|
| 381 |
|
| 382 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 383 |
def parse_and_generate_plot(df_data: list, indices_str: str):
|
| 384 |
"""Parse comma-separated indices and generate radar plot."""
|
| 385 |
if not indices_str or not indices_str.strip():
|
|
|
|
| 397 |
return generate_radar_plot([])
|
| 398 |
|
| 399 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 400 |
# Gradio UI
|
| 401 |
|
| 402 |
def build_app() -> gr.Blocks:
|
| 403 |
row_css = """
|
| 404 |
+
body { background-color: #f5f7fa !important; }
|
|
|
|
|
|
|
|
|
|
| 405 |
|
| 406 |
/* Row number column styling */
|
| 407 |
+
.metrics-table th:first-child, .metrics-table td:first-child {
|
| 408 |
+
width: 60px !important; text-align: center !important;
|
| 409 |
+
padding: 8px !important; font-weight: 600 !important;
|
|
|
|
|
|
|
|
|
|
| 410 |
background-color: #f0f0f0 !important;
|
| 411 |
}
|
|
|
|
| 412 |
.search-box {
|
| 413 |
+
background: white !important; padding: 16px !important;
|
| 414 |
+
border-radius: 6px; border: 2px solid #e1e4e8 !important;
|
| 415 |
+
box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06); margin-bottom: 16px;
|
|
|
|
|
|
|
|
|
|
| 416 |
}
|
| 417 |
+
.search-box .block { background: transparent !important; border: none !important; padding: 0 !important; }
|
| 418 |
+
.search-box label span { color: #24292e !important; font-weight: 600; font-size: 14px; margin-bottom: 8px; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 419 |
.search-box input.scroll-hide {
|
| 420 |
+
background-color: white !important; color: #24292e !important;
|
| 421 |
+
border: 1.5px solid #e1e4e8 !important; border-radius: 4px !important;
|
| 422 |
+
padding: 10px !important; box-shadow: none !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 423 |
}
|
| 424 |
+
.search-box input.scroll-hide:focus { border-color: #0366d6 !important; outline: none !important; }
|
| 425 |
|
| 426 |
+
.gradio-container { max-width: 100% !important; padding: 20px !important; background-color: #f5f7fa !important; }
|
| 427 |
+
.gradio-container .block, .gradio-container .form, .gradio-container .gr-box, .gradio-container .gr-input {
|
| 428 |
+
background-color: white !important; border-color: #e1e4e8 !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
| 429 |
}
|
| 430 |
+
.gradio-container label, .gradio-container p, .gradio-container span, .gradio-container div { color: #24292e !important; }
|
| 431 |
|
| 432 |
/* Table styling */
|
| 433 |
+
.gradio-container table.metrics-table th, .gradio-container table.metrics-table td {
|
| 434 |
+
padding: 10px 14px; border: 1.5px solid #e1e4e8; white-space: nowrap;
|
| 435 |
+
font-size: 13px; text-align: left; color: #24292e !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 436 |
}
|
|
|
|
| 437 |
.gradio-container table.metrics-table th {
|
| 438 |
background: linear-gradient(to bottom, #fafbfc, #f6f8fa);
|
| 439 |
+
font-weight: 600; position: sticky; top: 0; z-index: 10;
|
|
|
|
|
|
|
|
|
|
|
|
|
| 440 |
border-bottom: 2px solid #d1d5da;
|
| 441 |
}
|
| 442 |
+
.gradio-container table.metrics-table tbody tr:nth-child(even) { background-color: #f6f8fa; }
|
| 443 |
+
.gradio-container table.metrics-table tbody tr:hover { background-color: #e1e4e8; }
|
| 444 |
+
.gradio-container table.metrics-table { border-collapse: collapse; width: 100%; background: white; }
|
| 445 |
+
.gradio-container table.metrics-table a { color: #0366d6 !important; text-decoration: none; }
|
| 446 |
+
.gradio-container table.metrics-table a:hover { text-decoration: underline; }
|
| 447 |
|
| 448 |
+
/* Allow plot container to expand */
|
| 449 |
+
.gradio-container .plot-container { width: 100% !important; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 450 |
|
|
|
|
| 451 |
.table-container {
|
| 452 |
+
overflow-x: auto; overflow-y: auto; max-height: 75vh;
|
| 453 |
+
border: 2px solid #e1e4e8; border-radius: 6px;
|
| 454 |
+
background: white; box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
|
|
|
|
|
|
|
|
|
|
|
|
| 455 |
}
|
|
|
|
|
|
|
| 456 |
.filter-section {
|
| 457 |
+
background: white !important; padding: 0 !important; border-radius: 6px;
|
| 458 |
+
border: 2px solid #e1e4e8 !important; box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 459 |
}
|
| 460 |
+
.filter-section .wrap, .filter-section .block, .filter-section .container, .filter-section .group { background: transparent !important; }
|
| 461 |
+
.filter-section .wrap { padding: 20px !important; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 462 |
.gradio-container .accordion {
|
| 463 |
+
background: white !important; border: 2px solid #e1e4e8 !important;
|
| 464 |
+
border-radius: 6px !important; box-shadow: 0 1px 3px rgba(0, 0, 0, 0.06);
|
| 465 |
+
}
|
| 466 |
+
.info-section { padding: 16px; background: white !important; }
|
| 467 |
+
.info-section a { color: #0366d6 !important; }
|
| 468 |
+
.gradio-container h1 { color: #24292e !important; font-weight: 700; margin-bottom: 24px; }
|
| 469 |
+
.gradio-container h3 { color: #24292e !important; font-weight: 600; margin-bottom: 16px; }
|
| 470 |
+
.gradio-container input[type="checkbox"] { accent-color: #0366d6 !important; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 471 |
"""
|
| 472 |
|
|
|
|
| 473 |
with gr.Blocks(title="MoE-CAP Dashboard", css=row_css, theme=gr.themes.Default()) as demo:
|
| 474 |
gr.Markdown("# MoE-CAP Dashboard")
|
| 475 |
|
|
|
|
| 488 |
|
| 489 |
dir_path = gr.State(RESULT_DIR)
|
| 490 |
|
|
|
|
| 491 |
task_filter = gr.CheckboxGroup(
|
| 492 |
label="📊 Tasks",
|
| 493 |
choices=[
|
|
|
|
| 500 |
value=["gsm8k", "longbench", "mmlu", "numinamath", "ruler"]
|
| 501 |
)
|
| 502 |
|
|
|
|
| 503 |
framework_filter = gr.CheckboxGroup(
|
| 504 |
label="⚙️ Inference Frameworks",
|
| 505 |
choices=["sglang", "vllm"],
|
| 506 |
value=["sglang", "vllm"],
|
| 507 |
)
|
| 508 |
|
|
|
|
| 509 |
model_type_filter = gr.CheckboxGroup(
|
| 510 |
label="🤖 Model Types",
|
| 511 |
choices=["instruct", "thinking"],
|
| 512 |
value=["instruct", "thinking"],
|
| 513 |
)
|
| 514 |
|
|
|
|
| 515 |
precision_filter = gr.CheckboxGroup(
|
| 516 |
label="🎯 Precision",
|
| 517 |
choices=["bfloat16", "fp8"],
|
|
|
|
| 558 |
)
|
| 559 |
generate_btn = gr.Button("🎯 Generate", variant="primary", scale=1, size="lg")
|
| 560 |
|
| 561 |
+
# Modified Layout: Removed surrounding columns to allow plot to fill full width
|
| 562 |
+
radar_plot = gr.Plot(
|
| 563 |
+
label="",
|
| 564 |
+
value=generate_radar_plot([]),
|
| 565 |
+
elem_classes="plot-container"
|
| 566 |
+
)
|
|
|
|
| 567 |
|
| 568 |
df_data_state = gr.State([])
|
| 569 |
|