File size: 12,400 Bytes
bc37111 | 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 | """
Centralized Column Definitions
Single source of truth for all leaderboard columns.
Add new columns here and they propagate everywhere automatically.
"""
from dataclasses import dataclass
from enum import Enum, auto
from typing import List, Dict, Optional
class ColumnType(Enum):
"""Column data types for Gradio."""
NUMBER = "number"
STRING = "str"
HTML = "html"
class ColumnGroup(Enum):
"""Column groupings for organization and filtering."""
CORE = auto() # Always visible: Rank, Model
LEGAL = auto() # Legal benchmark scores
MTEB = auto() # MTEB task type scores
TOKENIZER = auto() # Tokenizer quality metrics
MODEL_INFO = auto() # Model metadata
CORRELATION = auto() # Correlation metrics
@dataclass
class ColumnDefinition:
"""
Complete definition for a leaderboard column.
This is the single source of truth - all column metadata lives here.
"""
name: str
api_name: Optional[str] = None
column_type: ColumnType = ColumnType.STRING
group: ColumnGroup = ColumnGroup.CORE
width: str = "120px"
decimals: int = 2
default_visible: bool = True
colorize: bool = False
description: str = ""
@property
def csv_key(self) -> str:
"""Get the key used in CSV files."""
return self.api_name or self.name
COLUMN_DEFINITIONS: List[ColumnDefinition] = [
# 1. Rank (always first)
ColumnDefinition(
name="Rank",
column_type=ColumnType.NUMBER,
group=ColumnGroup.CORE,
width="50px",
decimals=0,
default_visible=True,
description="Rank by MTEB Score (Mean TaskType)"
),
# 2. Model (always second)
ColumnDefinition(
name="Model",
column_type=ColumnType.HTML,
group=ColumnGroup.CORE,
width="280px",
default_visible=True,
colorize=False,
description="Model name with HuggingFace link"
),
# 3. MTEB Score - default
ColumnDefinition(
name="MTEB Score",
api_name="Mean (TaskType)",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="140px",
default_visible=True,
colorize=True,
description="MTEB Score: Average of task type category scores"
),
# 4. Legal Score - default
ColumnDefinition(
name="Legal Score",
api_name="Score(Legal)",
column_type=ColumnType.NUMBER,
group=ColumnGroup.LEGAL,
width="120px",
default_visible=True,
colorize=True,
description="Mean of legal benchmark scores (Contracts, Regulation, Caselaw)"
),
# 5. Pure Token Count - default
ColumnDefinition(
name="Pure Token Count",
column_type=ColumnType.NUMBER,
group=ColumnGroup.TOKENIZER,
width="150px",
decimals=0,
default_visible=True,
description="Tokens that are morphologically pure"
),
# 6. Max Sequence Length - default
ColumnDefinition(
name="Max Sequence Length",
api_name="Max Tokens",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MODEL_INFO,
width="160px",
decimals=0,
default_visible=True,
description="Maximum sequence length"
),
# 7. Parameters - default
ColumnDefinition(
name="Parameters",
api_name="Number of Parameters",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MODEL_INFO,
width="120px",
decimals=0,
default_visible=True,
description="Number of model parameters (e.g., 1.2B)"
),
# 8. Model Architecture - default
ColumnDefinition(
name="Model Architecture",
column_type=ColumnType.STRING,
group=ColumnGroup.MODEL_INFO,
width="180px",
default_visible=True,
description="Underlying model architecture (e.g., XLMRobertaModel)"
),
# 9. Mean (Task) - optional
ColumnDefinition(
name="Mean (Task)",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="120px",
default_visible=False,
colorize=True,
description="Average of all individual task scores"
),
# 10. Contracts - optional
ColumnDefinition(
name="Contracts",
column_type=ColumnType.NUMBER,
group=ColumnGroup.LEGAL,
width="110px",
default_visible=False,
colorize=True,
description="Performance on Turkish legal contract analysis"
),
# 11. Regulation - optional
ColumnDefinition(
name="Regulation",
column_type=ColumnType.NUMBER,
group=ColumnGroup.LEGAL,
width="110px",
default_visible=False,
colorize=True,
description="Performance on Turkish tax rulings retrieval"
),
# 12. Caselaw - optional
ColumnDefinition(
name="Caselaw",
column_type=ColumnType.NUMBER,
group=ColumnGroup.LEGAL,
width="110px",
default_visible=False,
colorize=True,
description="Performance on Court of Cassation case retrieval"
),
# 13. Classification - optional
ColumnDefinition(
name="Classification",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="130px",
default_visible=False,
colorize=True,
description="Performance on Turkish classification tasks"
),
# 14. Clustering - optional
ColumnDefinition(
name="Clustering",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="120px",
default_visible=False,
colorize=True,
description="Performance on Turkish clustering tasks"
),
# 15. Pair Classification - optional
ColumnDefinition(
name="Pair Classification",
api_name="PairClassification",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="150px",
default_visible=False,
colorize=True,
description="Performance on pair classification tasks (NLI)"
),
# 16. Retrieval - optional
ColumnDefinition(
name="Retrieval",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="120px",
default_visible=False,
colorize=True,
description="Performance on information retrieval tasks"
),
# 17. STS - optional
ColumnDefinition(
name="STS",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MTEB,
width="100px",
default_visible=False,
colorize=True,
description="Performance on Semantic Textual Similarity tasks"
),
# 18. Correlation - optional
ColumnDefinition(
name="Correlation",
column_type=ColumnType.NUMBER,
group=ColumnGroup.CORRELATION,
width="120px",
decimals=3,
default_visible=False,
colorize=True,
description="Weighted average of correlation metrics"
),
# 19. Tokenizer Type - optional
ColumnDefinition(
name="Tokenizer Type",
column_type=ColumnType.STRING,
group=ColumnGroup.TOKENIZER,
width="180px",
default_visible=False,
description="Tokenizer implementation type"
),
# 20. Unique Token Count - optional
ColumnDefinition(
name="Unique Token Count",
column_type=ColumnType.NUMBER,
group=ColumnGroup.TOKENIZER,
width="160px",
decimals=0,
default_visible=False,
description="Number of unique tokens on Turkish MMLU"
),
# 21. Turkish Token Count - optional
ColumnDefinition(
name="Turkish Token Count",
column_type=ColumnType.NUMBER,
group=ColumnGroup.TOKENIZER,
width="170px",
decimals=0,
default_visible=False,
description="Unique tokens that are valid Turkish"
),
# 22. Turkish Token % - optional
ColumnDefinition(
name="Turkish Token %",
column_type=ColumnType.NUMBER,
group=ColumnGroup.TOKENIZER,
width="140px",
default_visible=False,
description="Percentage of valid Turkish tokens"
),
# 23. Pure Token % - optional
ColumnDefinition(
name="Pure Token %",
column_type=ColumnType.NUMBER,
group=ColumnGroup.TOKENIZER,
width="130px",
default_visible=False,
description="Percentage of pure root word tokens"
),
# 24. Embed Dim - optional
ColumnDefinition(
name="Embed Dim",
api_name="Embedding Dimensions",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MODEL_INFO,
width="120px",
decimals=0,
default_visible=False,
description="Embedding dimension size"
),
# 25. Vocab Size - optional
ColumnDefinition(
name="Vocab Size",
column_type=ColumnType.NUMBER,
group=ColumnGroup.MODEL_INFO,
width="120px",
decimals=0,
default_visible=False,
description="Vocabulary size"
),
# 26. Model Type - optional
ColumnDefinition(
name="Model Type",
column_type=ColumnType.STRING,
group=ColumnGroup.MODEL_INFO,
width="130px",
default_visible=False,
description="Model type: Embedding, MLM, CLM-Embedding, or Seq2Seq"
),
]
class ColumnRegistry:
"""
Central registry for column definitions.
Provides convenient access methods for column metadata.
"""
def __init__(self, definitions: List[ColumnDefinition] = None):
self._definitions = definitions or COLUMN_DEFINITIONS
self._by_name: Dict[str, ColumnDefinition] = {
col.name: col for col in self._definitions
}
self._by_csv_key: Dict[str, ColumnDefinition] = {
col.csv_key: col for col in self._definitions
}
@property
def all_columns(self) -> List[str]:
"""All column names in order."""
return [col.name for col in self._definitions]
@property
def default_columns(self) -> List[str]:
"""Columns visible by default."""
return [col.name for col in self._definitions if col.default_visible]
@property
def optional_columns(self) -> List[str]:
"""Columns that can be toggled on/off."""
return [col.name for col in self._definitions if not col.default_visible]
@property
def score_columns(self) -> List[str]:
"""Columns that should be colorized."""
return [col.name for col in self._definitions if col.colorize]
@property
def numeric_columns(self) -> List[str]:
"""Columns with numeric type."""
return [col.name for col in self._definitions if col.column_type == ColumnType.NUMBER]
def get(self, name: str) -> Optional[ColumnDefinition]:
"""Get column definition by name."""
return self._by_name.get(name)
def get_by_csv_key(self, csv_key: str) -> Optional[ColumnDefinition]:
"""Get column definition by CSV key."""
return self._by_csv_key.get(csv_key)
def get_by_group(self, group: ColumnGroup) -> List[ColumnDefinition]:
"""Get all columns in a group."""
return [col for col in self._definitions if col.group == group]
def get_group_names(self, group: ColumnGroup) -> List[str]:
"""Get column names for a group."""
return [col.name for col in self.get_by_group(group)]
def get_datatypes(self, columns: List[str]) -> List[str]:
"""Get Gradio datatypes for given columns."""
return [
self._by_name[col].column_type.value
for col in columns
if col in self._by_name
]
def get_widths(self, columns: List[str]) -> List[str]:
"""Get column widths for given columns."""
return [
self._by_name[col].width
for col in columns
if col in self._by_name
]
def get_csv_mapping(self) -> Dict[str, str]:
"""Get mapping from CSV keys to display names."""
return {
col.csv_key: col.name
for col in self._definitions
if col.csv_key != col.name
}
# Global registry instance
column_registry = ColumnRegistry()
|