Spaces:
Running
Running
File size: 23,576 Bytes
fe79a14 |
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 |
# leaderboard/refresh.py
import json
import logging
from pathlib import Path
from typing import Any, Dict, List, Optional
import pandas as pd
import yaml
# --- Logging Setup ---
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(module)s - %(message)s"
)
logger = logging.getLogger(__name__)
# --- Path Definitions ---
SCRIPT_DIR = Path(__file__).resolve().parent
PROJECT_ROOT = SCRIPT_DIR.parent
# --- Default Input/Output Paths ---
DEFAULT_MODELS_FOLDER = PROJECT_ROOT.parent / "llm-leaderboard/models_info"
DEFAULT_RESULTS_FOLDER = PROJECT_ROOT.parent / "llm-leaderboard/results"
OUTPUT_FOLDER = SCRIPT_DIR / "boards_data"
CONFIG_FILE_PATH = SCRIPT_DIR / "leaderboard_config.yaml"
TEMPLATE_FOLDER = SCRIPT_DIR / "template_jsons"
# --- Constants for Subtask Processing ---
NLU_NLG_TASK_KEYS = ["persian_nlu", "persian_nlg"]
ALL_LEADERBOARD_COLUMNS = [
'Model Name', 'model_url', 'parameters_count', 'source_type', 'Average',
'Persian IFEval', 'Persian MT-Bench', "PerMMLU",
"PerCoR", "Persian NLU", "Persian NLG"
]
def load_tasks_from_config(config_path: Path) -> Dict[str, str]:
if not config_path.exists():
logger.error(f"Configuration file not found: {config_path}. Cannot load tasks.")
return {}
try:
with open(config_path, 'r', encoding='utf-8') as f:
config_data = yaml.safe_load(f)
tasks_from_config = config_data.get('task_display_names', {})
if not isinstance(tasks_from_config, dict):
logger.error(f"'task_display_names' in {config_path} is not a dictionary.")
return {}
processed_tasks = {k: v for k, v in tasks_from_config.items() if str(k).lower() != 'all'}
if not processed_tasks:
logger.warning(f"No tasks in {config_path} under 'task_display_names' (excluding 'all').")
return processed_tasks
except Exception as e:
logger.error(f"Error loading config {config_path}: {e}")
return {}
class ModelEvaluationProcessor:
def __init__(
self,
models_info_path: Path,
results_base_path: Path,
output_path: Path,
template_jsons_path: Path,
) -> None:
self.models_info_path = models_info_path
self.results_base_path = results_base_path
self.output_path = output_path
self.template_folder = template_jsons_path
self.output_path.mkdir(parents=True, exist_ok=True)
self.tasks_config = load_tasks_from_config(CONFIG_FILE_PATH)
if not self.tasks_config:
logger.error("Tasks config is empty. Processing might be affected.")
self.main_scores_map = {
"ifeval": "strict_instruction_accuracy",
"mt_bench": "score_mean",
"MMLU": "acc",
"persian_csr": "acc",
"persian_nlg": "nlg_score",
"persian_nlu": "nlu_score",
}
def _load_template(self, task_key: str) -> Dict[str, Any]:
path = self.template_folder / f"{task_key}.json"
try:
return json.loads(path.read_text(encoding="utf-8"))
except FileNotFoundError:
logger.warning(f"Template file not found for task_key {task_key} at {path}. Using empty template.")
return {}
except Exception as e:
logger.error(f"Cannot load template for task_key {task_key} from {path}: {e}")
return {}
def _deep_override(self, base: Any, override: Any) -> Any:
if isinstance(base, dict) and isinstance(override, dict):
merged = {}
for k, v_base in base.items():
if k in override and override[k] is not None and override[k] != -1:
merged[k] = self._deep_override(v_base, override[k])
else:
merged[k] = v_base
# for k, v_override in override.items():
# if k not in merged:
# merged[k] = v_override
return merged
elif override is not None and override != -1:
return override
else:
return base
def _load_model_raw_results(self, model_folder_name: str, task_key: str) -> Dict[str, Any]:
results_filename = f"{model_folder_name}___{task_key}.json"
results_file_path = self.results_base_path / results_filename
if results_file_path.exists():
try:
with open(results_file_path, 'r', encoding='utf-8') as f:
data = json.load(f)
return data if isinstance(data, dict) else {}
except json.JSONDecodeError as e:
logger.error(f"JSONDecodeError for model '{model_folder_name}', task_key '{task_key}' from {results_file_path}: {e}")
except Exception as e:
logger.error(f"Error loading results for model '{model_folder_name}', task_key '{task_key}' from {results_file_path}: {e}")
else:
logger.warning(f"Results file not found for model '{model_folder_name}', task_key '{task_key}' at {results_file_path}")
return {}
def load_and_fill_task_results(self, model_folder_name: str, task_key: str) -> Dict[str, Any]:
template = self._load_template(task_key)
raw_results = self._load_model_raw_results(model_folder_name, task_key)
return self._deep_override(template, raw_results)
def clean_previous_subtask_files(self) -> None:
logger.info("Cleaning previous NLU/NLG subtask JSONL files...")
for task_key_prefix in NLU_NLG_TASK_KEYS:
for result_file in self.results_base_path.glob(f"*___{task_key_prefix}.json"):
try:
task_data_content = result_file.read_text(encoding="utf-8")
if not task_data_content.strip():
logger.debug(f"Skipping empty result file for subtask cleaning: {result_file}")
continue
task_data = json.loads(task_data_content)
main_score_for_this_task_prefix = self.main_scores_map.get(task_key_prefix)
for subtask_name in task_data:
if subtask_name == main_score_for_this_task_prefix:
continue
if isinstance(task_data.get(subtask_name), dict):
subtask_output_path = self.output_path / f"{subtask_name}.jsonl"
if subtask_output_path.exists():
subtask_output_path.unlink()
logger.info(f"Deleted previous subtask file: {subtask_output_path}")
except json.JSONDecodeError as e:
logger.warning(f"Failed to decode JSON for subtask cleaning from {result_file}: {e}")
except Exception as e:
logger.warning(f"Failed to inspect/delete subtask files based on {result_file}: {e}")
def _process_subtask_data(self, task_results: Dict[str, Any], base_model_info: Dict[str, Any], parent_task_main_score_key: Optional[str], parent_task_key_for_log: str) -> None:
parent_task_main_score_value = task_results.get(parent_task_main_score_key) if parent_task_main_score_key else None
for subtask_name, subtask_scores_dict in task_results.items():
if subtask_name == parent_task_main_score_key:
continue
if not isinstance(subtask_scores_dict, dict):
logger.debug(f"Skipping entry '{subtask_name}' in '{parent_task_key_for_log}': not a dictionary of subtask scores.")
continue
row_data = base_model_info.copy()
row_data.update(subtask_scores_dict)
if parent_task_main_score_key:
row_data[parent_task_main_score_key] = parent_task_main_score_value
subtask_output_file = f"{subtask_name}.jsonl"
subtask_output_path = self.output_path / subtask_output_file
try:
current_entries = []
if subtask_output_path.exists():
existing_df = pd.read_json(subtask_output_path, lines=True)
if not existing_df.empty and 'Model Name' in existing_df.columns:
current_entries = existing_df[existing_df['Model Name'] != row_data['Model Name']].to_dict(orient='records')
current_entries.append(row_data)
updated_df = pd.DataFrame(current_entries)
updated_df.to_json(subtask_output_path, orient="records", lines=True, force_ascii=False)
logger.debug(f"Updated subtask file: {subtask_output_path} for model {base_model_info.get('Model Name')}, parent task {parent_task_key_for_log}")
except Exception as e:
logger.error(f"Error updating subtask file {subtask_output_path} for parent {parent_task_key_for_log}: {e}")
def process_nlu_nlg_subtasks(self, model_details: Dict[str, Any], model_folder_name: str, canonical_model_name: str) -> None:
common_subtask_model_info = {
"Model Name": canonical_model_name,
"model_url": model_details.get('model_url', model_details.get('link', model_details.get('homepage', 'https://google.com'))),
"parameters_count": str(model_details.get('n_parameters', "N/A")),
"source_type": "Closed-Source" # Default, will be refined
}
parameters_count_raw = model_details.get('n_parameters', None)
if parameters_count_raw is not None:
is_open_source_candidate = False
if isinstance(parameters_count_raw, (int, float)) and parameters_count_raw > 0:
is_open_source_candidate = True
elif isinstance(parameters_count_raw, str) and \
str(parameters_count_raw).strip().lower() not in ["", "n/a", "unknown", "private", "confidential", "tbd", "null", "closed"]:
is_open_source_candidate = True
common_subtask_model_info["source_type"] = "Open-Source" if is_open_source_candidate else "Closed-Source"
for task_key_for_subtasks in NLU_NLG_TASK_KEYS:
if task_key_for_subtasks not in self.tasks_config:
logger.debug(f"Subtask processing for '{task_key_for_subtasks}' skipped: not in tasks_config.")
continue
logger.info(f"Processing subtasks for '{task_key_for_subtasks}' for model '{canonical_model_name}'...")
parent_task_full_results = self.load_and_fill_task_results(model_folder_name, task_key_for_subtasks)
main_score_key_for_parent_task = self.main_scores_map.get(task_key_for_subtasks)
if not main_score_key_for_parent_task:
logger.warning(f"No main score key in main_scores_map for parent task '{task_key_for_subtasks}'.")
self._process_subtask_data(
parent_task_full_results,
common_subtask_model_info,
main_score_key_for_parent_task,
task_key_for_subtasks
)
def process_models(self) -> Dict[str, pd.DataFrame]:
processed_task_data: Dict[str, List[Dict[str, Any]]] = {task_key: [] for task_key in self.tasks_config.keys()}
all_models_summary_data: List[Dict[str, Any]] = []
if not self.models_info_path.exists() or not self.models_info_path.is_dir():
logger.critical(f"Configured MODELS_FOLDER path does not exist or is not a directory: {self.models_info_path}")
empty_dfs = {key: pd.DataFrame() for key in self.tasks_config.keys()}
empty_dfs["all"] = pd.DataFrame()
return empty_dfs
model_info_files = list(self.models_info_path.glob("*.json"))
if not model_info_files:
logger.warning(f"No model info files (*.json) found in {self.models_info_path}. No models will be processed.")
empty_dfs = {key: pd.DataFrame() for key in self.tasks_config.keys()}
empty_dfs["all"] = pd.DataFrame()
return empty_dfs
for model_info_file in model_info_files:
model_folder_name = model_info_file.stem
try:
with open(model_info_file, 'r', encoding='utf-8') as f:
model_details = json.load(f)
canonical_model_name = model_details.get('name_for_leaderboard',
model_details.get('model_hf_id',
model_details.get('name', model_folder_name)))
model_url = model_details.get('model_url', model_details.get('link', model_details.get('homepage', 'https_google.com')))
if not model_url: model_url = 'https_google.com'
parameters_count_raw = model_details.get('n_parameters', None)
parameters_count_display = str(parameters_count_raw) if parameters_count_raw is not None else "N/A"
source_type = "Closed-Source"
if parameters_count_raw is not None:
is_open_source_candidate = False
if isinstance(parameters_count_raw, (int, float)) and parameters_count_raw > 0:
is_open_source_candidate = True
elif isinstance(parameters_count_raw, str) and \
str(parameters_count_raw).strip().lower() not in ["", "n/a", "unknown", "private", "confidential", "tbd", "null", "closed"]:
is_open_source_candidate = True
source_type = "Open-Source" if is_open_source_candidate else "Closed-Source"
except Exception as e:
logger.error(f"Error loading/parsing model info from {model_info_file}: {e}. Skipping '{model_folder_name}'.")
continue
logger.info(f"Processing model: {canonical_model_name} (source ID: {model_folder_name})")
current_model_scores_for_summary: Dict[str, Any] = {
"Model Name": canonical_model_name,
"model_url": model_url,
"parameters_count": parameters_count_display,
"source_type": source_type
}
for task_key, task_display_name in self.tasks_config.items():
task_specific_results = self.load_and_fill_task_results(model_folder_name, task_key)
main_score_metric_name = self.main_scores_map.get(task_key)
task_data_entry_for_specific_jsonl: Dict[str, Any] = {
"Model Name": canonical_model_name,
"model_url": model_url,
"parameters_count": parameters_count_display,
"source_type": source_type
}
if isinstance(task_specific_results, dict) and task_specific_results:
for metric, value in task_specific_results.items():
task_data_entry_for_specific_jsonl[metric] = value
if main_score_metric_name and main_score_metric_name in task_specific_results:
score_value = task_specific_results[main_score_metric_name]
if task_key == "mt_bench" and score_value is not None:
try:
score_value = float(score_value) / 10.0
except (ValueError, TypeError):
logger.warning(f"Could not convert mt_bench score '{score_value}' to float for division for model {canonical_model_name}")
score_value = pd.NA
current_model_scores_for_summary[task_display_name] = score_value
elif main_score_metric_name:
logger.warning(f"Main score metric '{main_score_metric_name}' for task '{task_key}' (Display: {task_display_name}) not found for model '{canonical_model_name}'. Will be NA.")
current_model_scores_for_summary[task_display_name] = pd.NA
task_data_entry_for_specific_jsonl[main_score_metric_name] = pd.NA
else:
logger.warning(f"No valid results data for model '{canonical_model_name}', task_key '{task_key}'. Scores will be NA.")
if main_score_metric_name:
task_data_entry_for_specific_jsonl[main_score_metric_name] = pd.NA
current_model_scores_for_summary[task_display_name] = pd.NA
processed_task_data[task_key].append(task_data_entry_for_specific_jsonl)
all_models_summary_data.append(current_model_scores_for_summary)
self.process_nlu_nlg_subtasks(model_details, model_folder_name, canonical_model_name)
final_dataframes: Dict[str, pd.DataFrame] = {}
for task_key, data_list in processed_task_data.items():
df = pd.DataFrame(data_list) if data_list else pd.DataFrame()
main_score_col = self.main_scores_map.get(task_key)
if not df.empty and main_score_col and main_score_col in df.columns:
try:
df[main_score_col] = pd.to_numeric(df[main_score_col], errors='coerce')
# Sort by main score (NaNs will go last or first depending on na_position, default is last)
df = df.sort_values(by=main_score_col, ascending=False, na_position='last')
except Exception as e:
logger.warning(f"Could not sort dataframe for task {task_key} by score {main_score_col}: {e}")
final_dataframes[task_key] = df
if df.empty:
logger.warning(f"No data processed for task '{task_key}'. Resulting DataFrame is empty.")
if all_models_summary_data:
all_df = pd.DataFrame(all_models_summary_data)
score_cols_for_average = []
for _, task_display_name_for_avg in self.tasks_config.items():
if task_display_name_for_avg in all_df.columns:
numeric_col = pd.to_numeric(all_df[task_display_name_for_avg], errors='coerce')
if numeric_col.notna().any(): # Check if there is at least one non-NA numeric value
all_df[task_display_name_for_avg] = numeric_col
score_cols_for_average.append(task_display_name_for_avg)
else: # All values are NA or non-numeric
all_df[task_display_name_for_avg] = pd.NA # Ensure column is NA if not usable
logger.warning(f"Column '{task_display_name_for_avg}' for averaging in 'all' table is not numeric or all NaN. Excluding from average calculation and setting to NA.")
if score_cols_for_average:
try:
# Calculate mean; it will be NaN if any constituent score for a row is NaN.
all_df["Average"] = all_df[score_cols_for_average].mean(axis=1, skipna=False)
# Round only non-NaN averages
all_df.loc[all_df["Average"].notna(), "Average"] = all_df.loc[all_df["Average"].notna(), "Average"].round(4)
except Exception as e:
logger.error(f"Error calculating 'Average' for 'all' table: {e}. Average column might be NA or incorrect.")
all_df["Average"] = pd.NA # Fallback to NA
else:
logger.warning("No valid numeric score columns found to calculate 'Average' for 'all' table.")
all_df["Average"] = pd.NA # Assign pd.NA if no columns to average
# Sort 'all' table by Average (NaNs will be placed last by default with ascending=False)
if "Average" in all_df.columns: # Check if 'Average' column exists
# NaNs are typically sorted to the end by default when ascending=False or na_position='last'
all_df = all_df.sort_values(by="Average", ascending=False, na_position='last')
existing_cols_in_order = [col for col in ALL_LEADERBOARD_COLUMNS if col in all_df.columns]
other_cols = [col for col in all_df.columns if col not in existing_cols_in_order]
all_df = all_df[existing_cols_in_order + other_cols]
final_dataframes["all"] = all_df
else:
final_dataframes["all"] = pd.DataFrame()
logger.warning("No summary data collected for the 'all' table.")
return final_dataframes
def save_dataframe_as_jsonl(self, df: pd.DataFrame, filename_base: str) -> None:
if df is None or df.empty:
logger.warning(f"DataFrame for '{filename_base}.jsonl' is empty or None. Skipping save.")
return
output_file_path = self.output_path / f"{filename_base}.jsonl"
try:
df.to_json(output_file_path, orient="records", lines=True, force_ascii=False, index=False)
logger.info(f"Saved data to {output_file_path}")
except Exception as e:
logger.error(f"Failed to save DataFrame to {output_file_path}: {e}")
def run(self) -> None:
logger.info("Starting data processing pipeline in ModelEvaluationProcessor...")
self.clean_previous_subtask_files()
processed_dataframes = self.process_models()
for task_key_or_name, df in processed_dataframes.items():
self.save_dataframe_as_jsonl(df, task_key_or_name)
logger.info("Data processing pipeline completed successfully!")
def main() -> None:
models_folder_to_use = DEFAULT_MODELS_FOLDER
results_folder_to_use = DEFAULT_RESULTS_FOLDER
template_folder_to_use = TEMPLATE_FOLDER
logger.info(f"Refresh script running from: {SCRIPT_DIR}")
logger.info(f"CONFIGURED Input 'models_info' Path: {models_folder_to_use}")
logger.info(f"CONFIGURED Input 'results' Path: {results_folder_to_use}")
logger.info(f"CONFIGURED Input 'template_jsons' Path: {template_folder_to_use}")
logger.info(f"Outputting processed data to (inside 'leaderboard' dir): {OUTPUT_FOLDER}")
logger.info(f"Using configuration file (inside 'leaderboard' dir): {CONFIG_FILE_PATH}")
if not CONFIG_FILE_PATH.exists():
logger.critical(f"CRITICAL: Config file not found at {CONFIG_FILE_PATH}. Ensure '{CONFIG_FILE_PATH.name}' exists in '{SCRIPT_DIR}'.")
return
if not models_folder_to_use.exists() or not models_folder_to_use.is_dir():
logger.critical(f"CRITICAL: Input 'models_info' directory not found at {models_folder_to_use} or is not a directory.")
return
if not results_folder_to_use.exists() or not results_folder_to_use.is_dir():
logger.critical(f"CRITICAL: Input 'results' directory not found at {results_folder_to_use} or is not a directory.")
return
if not template_folder_to_use.exists() or not template_folder_to_use.is_dir():
logger.warning(f"WARNING: 'template_jsons' directory not found at {template_folder_to_use}. Template filling might not work as expected.")
try:
processor = ModelEvaluationProcessor(
models_info_path=models_folder_to_use,
results_base_path=results_folder_to_use,
output_path=OUTPUT_FOLDER,
template_jsons_path=template_folder_to_use,
)
processor.run()
except Exception as e:
logger.error(f"Unhandled exception in main: {e}", exc_info=True)
if __name__ == "__main__":
main() |