File size: 42,714 Bytes
3232d64 b1225dd 3232d64 b1225dd 3232d64 b1225dd 3232d64 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 |
import gradio as gr
import pandas as pd
import os
import sys
import traceback
import logging
# Disable SSL verification for curl requests if needed
os.environ['CURL_CA_BUNDLE'] = ''
# Configure minimal logging first thing - before any imports
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logging.getLogger("httpx").setLevel(logging.ERROR)
logging.getLogger("urllib3").setLevel(logging.ERROR)
logging.getLogger("matplotlib").setLevel(logging.WARNING)
logging.getLogger("huggingface_hub").setLevel(logging.ERROR)
from gradio.oauth import OAuthProfile
from src.display.about import (
CITATION_BUTTON_LABEL,
CITATION_BUTTON_TEXT,
EVALUATION_QUEUE_TEXT,
INTRODUCTION_TEXT,
LLM_BENCHMARKS_TEXT,
TITLE,
)
from src.display.css_html_js import custom_css
from src.utils import (
restart_space,
load_benchmark_results,
create_benchmark_plots,
create_combined_leaderboard_table,
create_evalmix_table,
create_light_eval_table,
create_raw_details_table,
create_human_arena_table,
update_supported_base_models
)
# Pipelines utils fonksiyonlarını import et
from pipelines.utils.common import search_and_filter
from pipelines.unified_benchmark import submit_unified_benchmark
# Evaluation types
EVAL_TYPES = ["EvalMix", "RAG-Judge", "Light-Eval", "Arena", "Snake-Bench"]
# Initialize OAuth configuration
OAUTH_CLIENT_ID = os.getenv("OAUTH_CLIENT_ID")
OAUTH_CLIENT_SECRET = os.getenv("OAUTH_CLIENT_SECRET")
OAUTH_SCOPES = os.getenv("OAUTH_SCOPES", "email")
OPENID_PROVIDER_URL = os.getenv("OPENID_PROVIDER_URL")
SESSION_TIMEOUT_MINUTES = int(os.getenv("HF_OAUTH_EXPIRATION_MINUTES", 30))
def format_dataframe(df, is_light_eval_detail=False):
"""
Float değerleri 2 ondalık basamağa yuvarla,
'file' sütununu kaldır ve kolon isimlerini düzgün formata getir
Args:
df: DataFrame to format
is_light_eval_detail: If True, use 4 decimal places for light eval detail results
"""
if df.empty:
return df
# 'file' sütununu kaldır
if 'file' in df.columns:
df = df.drop(columns=['file'])
# Specifically remove problematic columns
columns_to_remove = ["run_id", "user_id", "total_success_references", "Total Success References", "total_eval_samples",
"total_samples", "samples_number"]
for col in columns_to_remove:
if col in df.columns:
df = df.drop(columns=[col])
# Float değerleri yuvarlama - light eval detail için 4 hane, diğerleri için 2 hane
decimal_places = 4 if is_light_eval_detail else 2
for column in df.columns:
try:
if pd.api.types.is_float_dtype(df[column]):
df[column] = df[column].round(decimal_places)
except:
continue
# Kolon isimlerini düzgün formata getir
column_mapping = {}
for col in df.columns:
# Skip run_id and user_id fields
if col.lower() in ["run_id", "user_id"]:
continue
# Special handling for Turkish Semantic column
if "turkish_semantic" in col.lower():
column_mapping[col] = "Turkish Semantic"
continue
# Special handling for Multilingual Semantic column
if "multilingual_semantic" in col.lower():
column_mapping[col] = "Multilingual Semantic"
continue
# Skip already well-formatted columns or columns that contain special characters
if col == "Model Name" or " " in col:
# Still process column if it contains "mean"
if " mean" in col.lower():
cleaned_col = col.replace(" mean", "").replace(" Mean", "")
column_mapping[col] = cleaned_col
continue
# model_name column should be Model Name
if col == "model_name":
column_mapping[col] = "Model Name"
continue
# Remove the word "mean" from column names (case insensitive)
cleaned_col = col.replace(" mean", "").replace("_mean", "")
# Format column name by replacing underscores with spaces and capitalizing each word
formatted_col = " ".join([word.capitalize() for word in cleaned_col.replace("_", " ").split()])
column_mapping[col] = formatted_col
# Rename columns with the mapping
if column_mapping:
df = df.rename(columns=column_mapping)
return df
# User authentication function
def check_user_login(profile):
if profile is None:
return False, "Please log in with your Hugging Face account to submit models for benchmarking."
# In some environments, profile may be a string instead of a profile object
if isinstance(profile, str):
if profile == "":
return False, "Please log in with your Hugging Face account to submit models for benchmarking."
return True, f"Logged in as {profile}"
# Normal case where profile is an object with username attribute
return True, f"Logged in as {profile.username}"
def create_demo():
# Get logger for this function
logger = logging.getLogger("mezura")
with gr.Blocks(css=custom_css) as demo:
# Update supported base models at startup
logger.info("Updating supported base models at startup...")
update_supported_base_models()
logger.info("Base models updated successfully")
gr.Markdown(TITLE)
gr.Markdown(INTRODUCTION_TEXT)
# Hidden session state to track login expiration
session_expiry = gr.State(None)
try:
# Benchmark sonuçlarını yükle
benchmark_results = load_benchmark_results()
default_plots = create_benchmark_plots(benchmark_results, "avg")
# State variable to track login state across page refreshes
login_state = gr.State(value=False)
with gr.Tabs() as tabs:
with gr.TabItem("🏆 LLM Benchmark", elem_id="llm-benchmark-tab"):
gr.Markdown("## Model Evaluation Results")
gr.Markdown("This screen shows model performance across different evaluation categories.")
# Remove the separate refresh button row
# Instead, combine search and refresh in one row
with gr.Row():
search_input = gr.Textbox(
label="🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
placeholder="Enter model name or evaluation information...",
show_label=False
)
# # Update refresh button to be orange with "Refresh Results" text
# refresh_button = gr.Button("🔄 Refresh Results", variant="primary")
# # Status display for refresh results
# refresh_status = gr.Markdown("", visible=False)
# Benchmark tablarını semboller içeren tab grubuyla göster
with gr.Tabs() as benchmark_tabs:
with gr.TabItem("🏆 Leaderboard"):
# Birleşik leaderboard tablosu - avg_json dosyalarındaki tüm bilgileri göster
# Only use default data (avg files) for the leaderboard
combined_df = create_combined_leaderboard_table(benchmark_results)
# Float değerleri formatlama
combined_df = format_dataframe(combined_df)
# Tüm sütunları göster
if not combined_df.empty:
leaderboard_df = combined_df.copy()
else:
leaderboard_df = pd.DataFrame({"Model Name": ["No data available"]})
# Orijinal veriyi saklayacak state değişkeni
original_leaderboard_data = gr.State(value=leaderboard_df)
combined_table = gr.DataFrame(
value=leaderboard_df,
label="Model Performance Comparison",
interactive=False,
column_widths=["300px", "165px" ,"165px", "120px", "120px", "180px", "220px", "100px", "100px", "120px"]
)
with gr.TabItem("🏟️ Auto Arena"):
# Arena sonuçları - detail dosyalarını kullan
arena_details_df = create_raw_details_table(benchmark_results, "arena")
arena_details_df = format_dataframe(arena_details_df)
if arena_details_df.empty:
arena_details_df = pd.DataFrame({"model_name": ["No data available"]})
arena_table = gr.DataFrame(
value=arena_details_df,
label="Arena Detailed Results",
interactive=False,
column_widths=["300px", "150px", "110px", "110px", "180px", "100px", "120px"]
)
with gr.TabItem("👥 Human Arena"):
# Human Arena sonuçları - detail dosyalarını kullan
human_arena_data = benchmark_results["raw"]["human_arena"]
if human_arena_data:
human_arena_df = create_human_arena_table(human_arena_data)
else:
human_arena_df = pd.DataFrame()
human_arena_df = format_dataframe(human_arena_df)
if human_arena_df.empty:
human_arena_df = pd.DataFrame({"Model Name": ["No data available"]})
human_arena_table = gr.DataFrame(
value=human_arena_df,
label="Human Arena Results",
interactive=False,
column_widths=["300px", "150px", "110px", "110px", "110px", "156px", "169px", "100px", "120px"]
)
with gr.TabItem("📚 Retrieval"):
# RAG Judge sonuçları - detail dosyalarını kullan
rag_details_df = create_raw_details_table(benchmark_results, "retrieval")
rag_details_df = format_dataframe(rag_details_df)
if rag_details_df.empty:
rag_details_df = pd.DataFrame({"model_name": ["No data available"]})
rag_table = gr.DataFrame(
value=rag_details_df,
label="Retrieval Detailed Results",
interactive=False,
column_widths=["280px", "120px", "140px", "140px", "140px", "120px", "160px", "100px", "120px"]
)
with gr.TabItem("⚡ Light Eval"):
# Light Eval sonuçları - detail dosyalarını kullan
light_details_data = benchmark_results["raw"]["light_eval"]
if light_details_data:
light_details_df = create_light_eval_table(light_details_data, is_detail=True)
else:
light_details_df = pd.DataFrame()
light_details_df = format_dataframe(light_details_df, is_light_eval_detail=True)
if light_details_df.empty:
light_details_df = pd.DataFrame({"model_name": ["No data available"]})
light_table = gr.DataFrame(
value=light_details_df,
label="Light Eval Detailed Results",
interactive=False,
column_widths=["300px", "110px", "110px", "143px", "130px", "130px", "110px", "110px", "100px", "120px"]
)
with gr.TabItem("📋 EvalMix"):
# Hybrid Benchmark sonuçları - detail dosyalarını kullan
hybrid_details_df = create_raw_details_table(benchmark_results, "evalmix")
hybrid_details_df = format_dataframe(hybrid_details_df)
if hybrid_details_df.empty:
hybrid_details_df = pd.DataFrame({"model_name": ["No data available"]})
hybrid_table = gr.DataFrame(
value=hybrid_details_df,
label="EvalMix Detailed Results",
interactive=False,
column_widths=["300px", "180px", "230px", "143px", "110px", "110px", "110px", "110px", "169px", "220px" ,"100px", "120px"]
)
with gr.TabItem("🐍 𝐒𝐧𝐚𝐤𝐞 𝐁𝐞𝐧𝐜𝐡"):
# Snake Benchmark sonuçları - detail dosyalarını kullan
snake_details_df = create_raw_details_table(benchmark_results, "snake")
snake_details_df = format_dataframe(snake_details_df)
if snake_details_df.empty:
snake_details_df = pd.DataFrame({"model_name": ["No data available"]})
snake_table = gr.DataFrame(
value=snake_details_df,
label="Snake Benchmark Detailed Results",
interactive=False,
column_widths=["300px", "130px", "110px", "117px", "110px", "110px", "110px", "117px", "100px", "120px"]
)
# with gr.TabItem("📊 LM-Harness"):
# # LM Harness sonuçları - detail dosyalarını kullan
# lmharness_details_df = create_raw_details_table(benchmark_results, "lm_harness")
# lmharness_details_df = format_dataframe(lmharness_details_df)
#
# if lmharness_details_df.empty:
# lmharness_details_df = pd.DataFrame({"model_name": ["No data available"]})
#
# lmharness_table = gr.DataFrame(
# value=lmharness_details_df,
# label="LM Harness Detailed Results",
# interactive=False
# )
# # Refresh butonu bağlantısı
# refresh_button.click(
# refresh_leaderboard,
# inputs=[],
# outputs=[
# refresh_status,
# combined_table,
# hybrid_table,
# rag_table,
# light_table,
# arena_table,
# lmharness_table,
# snake_table
# ]
# )
# Tüm sekmeler için ortak arama fonksiyonu
def search_all_tabs(query, original_data):
"""
Tüm sekmelerde arama yapar
"""
if not query or query.strip() == "":
# Boş arama - orijinal veriyi döndür
return (original_data, arena_details_df, human_arena_df,
rag_details_df, light_details_df, hybrid_details_df, snake_details_df)
# Arama var - tüm sekmeleri filtrele
return (
search_and_filter(query, original_data, "All"),
search_and_filter(query, arena_details_df, "All"),
search_and_filter(query, human_arena_df, "All"),
search_and_filter(query, rag_details_df, "All"),
search_and_filter(query, light_details_df, "All"),
search_and_filter(query, hybrid_details_df, "All"),
search_and_filter(query, snake_details_df, "All")
)
# Arama fonksiyonu - tüm sekmeleri güncelle
search_input.change(
search_all_tabs,
inputs=[search_input, original_leaderboard_data],
outputs=[combined_table, arena_table, human_arena_table, rag_table, light_table, hybrid_table, snake_table]
)
with gr.TabItem("ℹ️ About", elem_id="about-tab"):
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
with gr.TabItem("📊 Datasets", elem_id="datasets-tab"):
gr.Markdown("## Benchmark Datasets")
gr.Markdown("""
This section provides detailed information about the datasets used in our evaluation benchmarks.
Each dataset has been carefully selected and adapted to provide comprehensive model evaluation across different domains and capabilities.
""")
# Create and display the datasets table
datasets_html = """
<div style="margin-top: 20px;">
<h3>Available Datasets for Evaluation</h3>
<table style="width: 100%; border-collapse: collapse; margin-top: 10px;">
<thead>
<tr style="background-color: var(--background-fill-secondary);">
<th style="padding: 12px; text-align: left; border-bottom: 2px solid var(--border-color-primary); width: 20%;">Dataset</th>
<th style="padding: 12px; text-align: left; border-bottom: 2px solid var(--border-color-primary); width: 18%;">Evaluation Task</th>
<th style="padding: 12px; text-align: left; border-bottom: 2px solid var(--border-color-primary); width: 10%;">Language</th>
<th style="padding: 12px; text-align: left; border-bottom: 2px solid var(--border-color-primary); width: 52%;">Description</th>
</tr>
</thead>
<tbody>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/mmlu_tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/mmlu_tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval MMLU</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish adaptation of MMLU (Massive Multitask Language Understanding) v0.2 covering 57 academic subjects including mathematics, physics, chemistry, biology, history, law, and computer science. Tests knowledge and reasoning capabilities across multiple domains with multiple-choice questions.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/truthful_qa-tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/truthful_qa-tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval TruthfulQA</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish version of TruthfulQA (v0.2) designed to measure model truthfulness and resistance to generating false information. Contains questions where humans often answer incorrectly due to misconceptions or false beliefs, testing the model's ability to provide accurate information.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/winogrande-tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/winogrande-tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval WinoGrande</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish adaptation of WinoGrande (v0.2) focusing on commonsense reasoning through pronoun resolution tasks. Tests the model's ability to understand context, make logical inferences, and resolve ambiguous pronouns in everyday scenarios.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/hellaswag_tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/hellaswag_tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval HellaSwag</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish version of HellaSwag (v0.2) for commonsense reasoning evaluation. Tests the model's ability to predict plausible continuations of everyday scenarios and activities, requiring understanding of common sense and typical human behavior patterns.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/arc-tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/arc-tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval ARC</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish adaptation of ARC (AI2 Reasoning Challenge) v0.2 focusing on science reasoning and question answering. Contains grade school level science questions that require reasoning beyond simple factual recall, covering topics in physics, chemistry, biology, and earth science.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/malhajar/gsm8k_tr-v0.2" target="_blank" style="color: #0066cc; text-decoration: none;">malhajar/gsm8k_tr-v0.2</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Lighteval GSM8K</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">Turkish version of GSM8K (Grade School Math 8K) v0.2 for mathematical reasoning evaluation. Contains grade school level math word problems that require multi-step reasoning, arithmetic operations, and logical problem-solving skills to arrive at the correct numerical answer.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/newmindai/mezura-eval-data" target="_blank" style="color: #0066cc; text-decoration: none;">newmindai/mezura-eval-data</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Auto-Arena</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">mezura-eval dataset is a Turkish-language legal text dataset designed for evaluation tasks with RAG context support. The subsets include domains like Environmental Law, Tax Law, Data Protection Law and Health Law each containing annotated samples. Every row includes structured fields such as the category, concept, input and contextual information drawn from sources like official decisions.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/newmindai/mezura-eval-data" target="_blank" style="color: #0066cc; text-decoration: none;">newmindai/mezura-eval-data</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">EvalMix</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">mezura-eval dataset is a Turkish-language legal text dataset designed for evaluation tasks with RAG context support. The subsets include domains like Environmental Law, Tax Law, Data Protection Law and Health Law each containing annotated samples. Every row includes structured fields such as the category, concept, input and contextual information drawn from sources like official decisions.</td>
</tr>
<tr>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 20%;"><a href="https://huggingface.co/datasets/newmindai/mezura-eval-data" target="_blank" style="color: #0066cc; text-decoration: none;">newmindai/mezura-eval-data</a></td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 18%;">Retrieval</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 10%;">Turkish</td>
<td style="padding: 10px; border-bottom: 1px solid var(--border-color-primary); width: 52%;">mezura-eval dataset is a Turkish-language legal text dataset designed for evaluation tasks with RAG context support. The subsets include domains like Environmental Law, Tax Law, Data Protection Law and Health Law each containing annotated samples. Every row includes structured fields such as the category, concept, input and contextual information drawn from sources like official decisions.</td>
</tr>
</tbody>
</table>
</div>
"""
gr.HTML(datasets_html)
with gr.TabItem("🔬 Evaluation", elem_id="evaluation-tab"):
gr.Markdown("""
<h2 align="center">Model Evaluation</h2>
### Evaluation Process:
1. **Login to Your Hugging Face Account**
- You must be logged in to submit models for evaluation
2. **Enter Model Name**
- Input the HuggingFace model name or path you want to evaluate
- Example: meta-llama/Meta-Llama-3.1-70B-Instruct
3. **Select Base Model**
- Choose the base model from the dropdown list
- The system will verify if your repository is a valid HuggingFace repository
- It will check if the model is trained from the selected base model
4. **Start Evaluation**
- Click the "Start All Benchmarks" button to begin the evaluation
- If validation passes, your request will be processed
- If validation fails, you'll see an error message
### Important Limitations:
- The model repository must be a maximum of 750 MB in size.
- For trained adapters, the maximum LoRA rank must be 32.
""")
# Authentication Component (Always visible)
auth_container = gr.Group()
with auth_container:
# Simplified login button - Gradio will handle the OAuth
login_button = gr.LoginButton()
# Get base models from API
from api.config import get_base_model_list
BASE_MODELS = get_base_model_list()
# Fallback to static list if API list is empty
if not BASE_MODELS:
BASE_MODELS = [
"meta-llama/Meta-Llama-3.1-8B-Instruct",
"meta-llama/Llama-3.2-3B-Instruct",
"meta-llama/Llama-3.3-70B-Instruc",
"Qwen/Qwen2.5-72B-Instruct",
"Qwen/QwQ-32B",
"google/gemma-2-2b-it"
]
# Content that's only visible when logged in
login_dependent_content = gr.Group(visible=False)
with login_dependent_content:
gr.Markdown("### Model Submission")
# Model input
model_to_evaluate = gr.Textbox(
label="Adapter Repo ID",
placeholder="e.g., valadapt/llama-3-8b-turkish"
)
# Add note about supported model types
gr.Markdown("""
**Note:** Currently, only adapter models are supported. Merged models are not yet supported.
""", elem_classes=["info-text"])
# Base model selection
base_model_dropdown = gr.Dropdown(
choices=BASE_MODELS,
label="Base Model",
allow_custom_value=True
)
# Reasoning capability checkbox
reasoning_checkbox = gr.Checkbox(
label="Reasoning",
value=False,
info="Enable reasoning capability during evaluation"
)
# Email input
email_input = gr.Textbox(
label="Email Address",
placeholder="example@domain.com",
info="You'll receive notification when benchmark is complete"
)
# Submit button - CRITICAL: This submit button is only visible when logged in
submit_button = gr.Button("Start All Benchmarks", variant="primary")
# Result area (initially empty)
result_output = gr.Markdown("")
# Status area for authentication errors (initially hidden)
auth_error = gr.Markdown(visible=False)
# Function to handle login visibility
def toggle_form_visibility(profile):
# User is not logged in
if profile is None:
return (
gr.update(visible=False),
gr.update(
visible=True,
value="<p style='color: red; text-align: center; font-weight: bold;'>Authentication required. Please log in with your Hugging Face account to submit models.</p>"
)
)
# Log successful authentication
try:
if hasattr(profile, 'name'):
username = profile.name
elif hasattr(profile, 'username'):
username = profile.username
else:
username = str(profile)
logger.info(f"User authenticated: {username}")
except Exception as e:
logger.info(f"LOGIN - Error inspecting profile: {str(e)}")
# User is logged in - show form, hide error
return (
gr.update(visible=True),
gr.update(visible=False, value="")
)
# Connect login button to visibility toggle
login_button.click(
fn=toggle_form_visibility,
inputs=[login_button],
outputs=[login_dependent_content, auth_error]
)
# Check visibility on page load
demo.load(
fn=toggle_form_visibility,
inputs=[login_button],
outputs=[login_dependent_content, auth_error]
)
# Handle submission with authentication check
def submit_model(model, base_model, reasoning, email, profile):
# Authentication check
if profile is None:
logging.warning("Unauthorized submission attempt with no profile")
return "<p style='color: red; font-weight: bold;'>Authentication required. Please log in with your Hugging Face account.</p>"
# IMPORTANT: In local development, Gradio returns "Sign in with Hugging Face" string
# This is NOT a real authentication, just a placeholder for local testing
if isinstance(profile, str) and profile == "Sign in with Hugging Face":
# Block submission in local dev with mock auth
return "<p style='color: orange; font-weight: bold;'>⚠️ HF authentication required.</p>"
# Email is required
if not email or email.strip() == "":
return "<p style='color: red; font-weight: bold;'>Email address is required to receive benchmark results.</p>"
# Check if the model is a merged model (not supported)
try:
from src.submission.check_validity import determine_model_type
model_type, _ = determine_model_type(model)
if model_type == "merged_model" or model_type == "merge":
return "<p style='color: red; font-weight: bold;'>Merged models are not supported yet. Please submit an adapter model instead.</p>"
except Exception as e:
# If error checking model type, continue with submission
logging.warning(f"Error checking model type: {str(e)}")
# Call the benchmark function with profile information
# base_model validasyonunu kaldırdık ama parametre olarak yine de gönderiyoruz
result_message, _ = submit_unified_benchmark(model, base_model, reasoning, email, profile)
logging.info(f"Submission processed for model: {model}")
return result_message
# Connect submit button
submit_button.click(
fn=submit_model,
inputs=[
model_to_evaluate,
base_model_dropdown,
reasoning_checkbox,
email_input,
login_button
],
outputs=[result_output]
)
except Exception as e:
traceback.print_exc()
gr.Markdown(f"## Error: An issue occurred while loading the LLM Benchmark screen")
gr.Markdown(f"Error message: {str(e)}")
gr.Markdown("Please check your configuration and try again.")
# Citation information at the bottom
gr.Markdown("---")
with gr.Accordion(CITATION_BUTTON_LABEL, open=False):
gr.Textbox(
value=CITATION_BUTTON_TEXT,
lines=10,
show_copy_button=True,
label=None
)
return demo
if __name__ == "__main__":
# Get app logger
logger = logging.getLogger("mezura")
# Additional sensitive filter for remaining logs
class SensitiveFilter(logging.Filter):
def filter(self, record):
msg = record.getMessage().lower()
# Filter out messages with tokens, URLs with sign= in them, etc
sensitive_patterns = ["token", "__sign=", "request", "auth", "http request"]
return not any(pattern in msg.lower() for pattern in sensitive_patterns)
# Apply the filter to all loggers
for logger_name in logging.root.manager.loggerDict:
logging.getLogger(logger_name).addFilter(SensitiveFilter())
try:
logger.info("Creating demo...")
demo = create_demo()
logger.info("Launching demo on 0.0.0.0...")
# Add options to fix the session.pop error
demo.launch(
server_name="0.0.0.0",
server_port=7860
)
except FileNotFoundError as e:
logger.critical(f"Configuration file not found: {e}")
print(f"\n\nERROR: Configuration file not found. Please ensure config/api_config.yaml exists.\n{e}\n")
sys.exit(1)
except ValueError as e:
logger.critical(f"Configuration error: {e}")
print(f"\n\nERROR: Invalid configuration. Please check your config/api_config.yaml file.\n{e}\n")
sys.exit(1)
except Exception as e:
logger.critical(f"Could not launch demo: {e}", exc_info=True)
|