Spaces:
Configuration error
Configuration error
File size: 11,527 Bytes
af2c3f6 | 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 | """
Evaluation Agent - Scores speech and translation models
"""
import numpy as np
from datetime import datetime, timezone
from typing import List, Dict, Optional
from src.models.schemas import (
ModelMetadata, UserRequirements, HardwareConstraint,
ModelScore, TaskType
)
class EvaluationAgent:
"""
Evaluates and scores translation, TTS, and STT models.
"""
def __init__(self, weights: Dict[str, float] = None):
self.weights = weights or {
'downloads': 0.20,
'recency': 0.15,
'license': 0.10,
'size': 0.15,
'performance': 0.25,
'language_match': 0.15
}
def score_models(self,
models: List[ModelMetadata],
requirements: UserRequirements) -> List[ModelScore]:
"""Score models based on multiple criteria"""
scored_models = []
for model in models:
scores = {}
# Common scores
scores['downloads'] = self._score_downloads(model.downloads)
scores['recency'] = self._score_recency(model.last_modified)
scores['license'] = self._score_license(model.license)
scores['size'] = self._score_size(model.model_size, requirements)
scores['performance'] = self._score_performance(model, requirements)
scores['language_match'] = self._score_language_match(model, requirements)
# Task-specific scores
if model.task_type in [TaskType.TEXT_GENERATION, TaskType.CHAT,
TaskType.INSTRUCTION_FOLLOWING, TaskType.CODE_GENERATION,
TaskType.QUESTION_ANSWERING, TaskType.SUMMARIZATION]:
scores['llm_capabilities'] = self._score_llm_capabilities(model, requirements)
elif model.task_type in [TaskType.OCR, TaskType.DOCUMENT_UNDERSTANDING]:
scores['ocr_capabilities'] = self._score_ocr_capabilities(model, requirements)
# Calculate weighted total
total_score = 0.0
weight_sum = 0.0
all_weights = {
'downloads': 0.15,
'recency': 0.10,
'license': 0.05,
'size': 0.10,
'performance': 0.20,
'language_match': 0.15,
'llm_capabilities': 0.25,
'ocr_capabilities': 0.25
}
for metric, score in scores.items():
if metric in all_weights:
total_score += score * all_weights[metric]
weight_sum += all_weights[metric]
if weight_sum > 0:
total_score /= weight_sum
# Apply hardware penalty
hardware_penalty = self._check_hardware_constraints(model, requirements)
total_score *= hardware_penalty
scored_models.append(ModelScore(
model_id=model.model_id,
task_type=model.task_type,
total_score=float(total_score),
component_scores=scores,
metadata=model
))
return sorted(scored_models, key=lambda x: x.total_score, reverse=True)
def _score_downloads(self, downloads: int) -> float:
"""Score based on downloads (log scale)"""
if downloads <= 0:
return 0.0
log_downloads = np.log10(downloads + 1)
return min(log_downloads / 6.0, 1.0)
def _score_recency(self, last_modified) -> float:
"""Score based on recency - FIXED timezone issue"""
if not last_modified:
return 0.5
try:
# Make last_modified timezone-naive for comparison
if hasattr(last_modified, 'tzinfo') and last_modified.tzinfo is not None:
# Convert to timezone-naive by removing timezone info
last_modified = last_modified.replace(tzinfo=None)
# Get current time as timezone-naive
now = datetime.now()
# Calculate days difference
days_since_update = (now - last_modified).days
if days_since_update < 30:
return 1.0
elif days_since_update < 90:
return 0.8
elif days_since_update < 180:
return 0.6
elif days_since_update < 365:
return 0.4
else:
return 0.2
except Exception as e:
print(f" Warning: Error calculating recency: {e}")
return 0.5
def _score_license(self, license: str) -> float:
"""Score based on license"""
license_lower = license.lower()
open_licenses = ['mit', 'apache', 'bsd', 'cc', 'gpl', 'lgpl']
if any(open_license in license_lower for open_license in open_licenses):
return 0.9
elif 'commercial' in license_lower:
return 0.5
else:
return 0.7
def _score_size(self, model_size: Optional[float], requirements: UserRequirements) -> float:
"""Score based on size - smaller is better"""
if not model_size:
return 0.5
if requirements.max_model_size_gb:
if model_size > requirements.max_model_size_gb:
return 0.0
size_ratio = 1.0 - (model_size / requirements.max_model_size_gb)
return 0.5 + (size_ratio * 0.5)
# No constraint - smaller is better
if model_size < 0.5:
return 1.0
elif model_size < 1.0:
return 0.9
elif model_size < 2.0:
return 0.7
elif model_size < 5.0:
return 0.5
else:
return 0.3
def _score_performance(self, model: ModelMetadata, requirements: UserRequirements) -> float:
"""Score based on performance metrics"""
metrics = model.performance_metrics
if not metrics:
return 0.5
if model.task_type == TaskType.TRANSLATION:
# Prefer BLEU scores
if 'bleu' in metrics:
return min(metrics['bleu'] / 50, 1.0) # BLEU up to 50
return 0.5
elif model.task_type == TaskType.SPEECH_TO_TEXT:
# Prefer low WER
if 'wer' in metrics:
return max(0, 1.0 - (metrics['wer'] / 100))
return 0.5
elif model.task_type == TaskType.TEXT_TO_SPEECH:
# Prefer more voices and higher sample rate
score = 0.5
if model.voice_count > 0:
score += min(model.voice_count / 10, 0.3)
if model.sample_rate and model.sample_rate >= 16000:
score += 0.2
return min(score, 1.0)
return 0.5
def _score_language_match(self, model: ModelMetadata, requirements: UserRequirements) -> float:
"""Score based on language support"""
if requirements.task_type == TaskType.TRANSLATION and requirements.translation_reqs:
req = requirements.translation_reqs
source = req.source_language.value
target = req.target_language.value
score = 0.5
if model.source_languages and source in model.source_languages:
score += 0.25
if model.target_languages and target in model.target_languages:
score += 0.25
return score
elif requirements.task_type == TaskType.TEXT_TO_SPEECH and requirements.tts_reqs:
req = requirements.tts_reqs
lang = req.language.value
if model.languages and lang in model.languages:
return 1.0
return 0.5
elif requirements.task_type == TaskType.SPEECH_TO_TEXT and requirements.stt_reqs:
req = requirements.stt_reqs
lang = req.language.value
if model.languages and lang in model.languages:
return 1.0
return 0.5
return 0.5
def _score_llm_capabilities(self, model: ModelMetadata, requirements: UserRequirements) -> float:
"""Score LLM based on capabilities"""
if not requirements.llm_reqs:
return 0.5
req = requirements.llm_reqs
score = 0.5
# Check context length
if model.context_length and req.context_length:
if model.context_length >= req.context_length:
score += 0.2
else:
score -= 0.1
# Check chat template
if req.wants_chat_template and model.has_chat_template:
score += 0.2
# Check function calling
if req.wants_function_calling and model.supports_function_calling:
score += 0.2
# Check code generation
if req.wants_code_generation and model.supports_code:
score += 0.2
# Check instruction following
if req.wants_instruction_following and model.supports_instruction:
score += 0.2
# Check quantization support
if req.quantization and req.quantization in model.quantization_supported:
score += 0.1
return min(score, 1.0)
def _score_ocr_capabilities(self, model: ModelMetadata, requirements: UserRequirements) -> float:
"""Score OCR model based on capabilities"""
if not requirements.ocr_reqs:
return 0.5
req = requirements.ocr_reqs
score = 0.5
# Check handwriting support
if req.handwritten and model.supports_handwriting:
score += 0.3
elif req.handwritten and not model.supports_handwriting:
score -= 0.2
# Check layout analysis
if req.wants_layout_analysis and model.supports_layout:
score += 0.2
# Check table extraction
if req.wants_table_extraction and model.supports_tables:
score += 0.2
# Check formula recognition
if req.wants_formula_recognition and model.supports_formulas:
score += 0.2
return min(score, 1.0)
def _check_hardware_constraints(self, model: ModelMetadata,
requirements: UserRequirements) -> float:
"""Apply penalty if hardware constraints not met"""
if not requirements.hardware_constraints:
return 1.0
hardware = model.hardware_requirements
for constraint in requirements.hardware_constraints:
if constraint == HardwareConstraint.CPU:
if hardware.get('cpu_compatible', True):
return 1.0
elif constraint in [HardwareConstraint.GPU_4GB, HardwareConstraint.GPU_8GB,
HardwareConstraint.GPU_16GB, HardwareConstraint.GPU_24GB,
HardwareConstraint.GPU_40GB, HardwareConstraint.GPU_80GB]:
if hardware.get('gpu_required', False):
return 0.9 # Small penalty for requiring GPU
return 1.0
return 0.8 |