Spaces:
Running
Running
File size: 17,171 Bytes
61a383e 56a8632 d7fb7f8 56a8632 61a383e c27741d 0209035 7d60219 0209035 c27741d 61a383e 56a8632 61a383e 75bf67b 61a383e 56a8632 61a383e c27741d 56a8632 61a383e 75bf67b 0209035 56a8632 75bf67b c27741d 56a8632 c27741d 56a8632 c27741d 56a8632 c27741d 56a8632 c27741d 7d60219 c27741d 56a8632 7d60219 56a8632 c27741d 56a8632 75bf67b da5f99b 75bf67b da5f99b 75bf67b 7d60219 da5f99b 6ea2d23 3d03815 da5f99b 75bf67b da5f99b 5a536b8 da5f99b 75bf67b da5f99b 7d60219 da5f99b 4adde24 da5f99b 5a536b8 da5f99b 75bf67b 61a383e 75bf67b 61a383e 75bf67b 61a383e 75bf67b 7531f8a 61a383e 5a536b8 61a383e 7531f8a 61a383e 7531f8a 61a383e 75bf67b 61a383e 75bf67b 61a383e 75bf67b 61a383e 75bf67b 61a383e 56a8632 61a383e 2e6aca3 61a383e 2e6aca3 56a8632 2e6aca3 56a8632 61a383e 75bf67b 61a383e 56a8632 61a383e 56a8632 61a383e 75bf67b 61a383e 56a8632 75bf67b 61a383e 75bf67b 56a8632 75bf67b 56a8632 61a383e 2e6aca3 61a383e 2e6aca3 61a383e 2457f4e 2e6aca3 2457f4e 2e6aca3 2457f4e 2e6aca3 61a383e 2e6aca3 9b0f151 |
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 |
"""
Module for evaluation and annotation of bot responses
"""
import json
import os
import datetime
from typing import List, Dict, Any, Tuple, Optional
import io
import logging
from huggingface_hub import HfApi
logger = logging.getLogger(__name__)
from config.settings import (
DATASET_ID,
HF_TOKEN,
CHAT_HISTORY_PATH,
DATASET_CHAT_HISTORY_PATH,
DATASET_ANNOTATIONS_PATH
)
class ChatEvaluator:
def __init__(self, hf_token: str = None, dataset_id: str = None):
"""
Initialize chat evaluator with lazy loading
Args:
hf_token: Hugging Face token
dataset_id: Dataset ID on Hugging Face
"""
self.hf_token = hf_token or HF_TOKEN
self.dataset_id = dataset_id or DATASET_ID
self.api = HfApi(token=self.hf_token)
# Using paths from settings
self.chat_history_path = DATASET_CHAT_HISTORY_PATH
self.annotations_path = DATASET_ANNOTATIONS_PATH
# Cache for chat histories and QA pairs
self._chat_histories = None
self._qa_pairs = None
self._annotations = None
# Ensure directories exist in dataset
try:
self._ensure_dataset_structure()
except Exception as e:
logger.error(f"Failed to ensure dataset structure: {e}")
def _ensure_dataset_structure(self):
"""Ensure required directories exist in dataset"""
try:
files = self.api.list_repo_files(self.dataset_id, repo_type="dataset")
# Check and create chat history directory
if self.chat_history_path not in files:
self.api.upload_file(
path_or_fileobj=io.BytesIO(b""),
path_in_repo=f"{self.chat_history_path}/.gitkeep",
repo_id=self.dataset_id,
repo_type="dataset"
)
# Check and create annotations directory
if self.annotations_path not in files:
self.api.upload_file(
path_or_fileobj=io.BytesIO(b""),
path_in_repo=f"{self.annotations_path}/.gitkeep",
repo_id=self.dataset_id,
repo_type="dataset"
)
except Exception as e:
logger.error(f"Error ensuring dataset structure: {e}")
raise
def reset_cache(self):
"""
Reset the cache to force reload of data
"""
self._chat_histories = None
self._qa_pairs = None
self._annotations = None
logger.info("Chat evaluator cache has been reset")
def get_chat_history(self, force_reload=False) -> List[Dict[str, Any]]:
"""
Get all chat histories from the dataset
Args:
force_reload: If True, ignore cache and reload from dataset
"""
# Return cached data if available and not forcing reload
if self._chat_histories is not None and not force_reload:
logger.debug("Returning cached chat histories")
return self._chat_histories
try:
# Get list of all files in chat history directory
files = self.api.list_repo_files(self.dataset_id, repo_type="dataset")
# Filter for chat history files
chat_path = f"{self.chat_history_path}/"
chat_files = [f for f in files if f.startswith(chat_path) and f.endswith('.json')]
logger.debug(f"Found {len(chat_files)} chat files") # More compact log
histories = []
for file in chat_files:
try:
# Download and parse each chat file
content = self.api.hf_hub_download(
repo_id=self.dataset_id,
filename=file,
repo_type="dataset"
)
with open(content, 'r', encoding='utf-8') as f:
chat_data = json.load(f)
if isinstance(chat_data, dict) and 'history' in chat_data:
histories.append(chat_data)
else:
logger.warning(f"Invalid chat history format in {file}")
except Exception as e:
logger.error(f"Error processing chat file {file}: {e}")
continue
# Cache the results
self._chat_histories = histories
return histories
except Exception as e:
logger.error(f"Failed to get chat histories: {e}")
return []
def extract_qa_pairs(self, histories: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
"""
Extract question-answer pairs from chat histories
"""
qa_pairs = []
for history in histories:
messages = history.get('history', [])
current_question = None
for msg in messages:
if msg.get('role') == 'user':
current_question = msg.get('content')
elif msg.get('role') == 'assistant' and current_question:
qa_pairs.append({
'conversation_id': history.get('conversation_id'),
'question': current_question,
'answer': msg.get('content'),
'timestamp': history.get('timestamp')
})
current_question = None
logger.debug(f"Extracted {len(qa_pairs)} QA pairs")
return qa_pairs
def get_qa_pairs_for_evaluation(self, limit: int = 50, force_reload=False) -> List[Dict[str, Any]]:
"""
Extract question-answer pairs for evaluation
Args:
limit: Maximum number of pairs to return
force_reload: If True, force reload from dataset
Returns:
List of QA pairs with metadata
"""
# Return cached data if available and not forcing reload
if self._qa_pairs is not None and not force_reload:
logger.debug("Returning cached QA pairs")
return self._qa_pairs[:limit] # Respect the limit parameter
chat_data = self.get_chat_history(force_reload=force_reload)
qa_pairs = []
logger.debug(f"Processing {len(chat_data)} chat histories")
for chat in chat_data:
conversation_id = chat.get("conversation_id", "unknown")
timestamp = chat.get("timestamp", "")
messages = chat.get("history", [])
# Find user-assistant pairs in messages
for i in range(len(messages) - 1):
if (messages[i].get("role") == "user" and
messages[i+1].get("role") == "assistant"):
question = messages[i].get("content", "").strip()
answer = messages[i+1].get("content", "").strip()
# Only include non-empty pairs
if question and answer:
qa_pairs.append({
"conversation_id": conversation_id,
"timestamp": timestamp,
"question": question,
"original_answer": answer,
"question_timestamp": messages[i].get("timestamp", ""),
"answer_timestamp": messages[i+1].get("timestamp", "")
})
# Cache the results
self._qa_pairs = qa_pairs
logger.debug(f"Extracted {len(qa_pairs)} QA pairs")
# Return up to the limit
return qa_pairs[:limit]
def get_evaluation_status(self, force_reload=False) -> Dict[str, int]:
"""
Get status of evaluated QA pairs
Args:
force_reload: If True, force reload from dataset
Returns:
Dictionary with counts of evaluated and unevaluated QA pairs
"""
all_pairs = self.get_qa_pairs_for_evaluation(limit=1000, force_reload=force_reload) # Get a large sample
evaluated_pairs = self.get_annotations(force_reload=force_reload)
# Count evaluated conversation IDs
evaluated_ids = set(item.get("conversation_id") for item in evaluated_pairs)
return {
"total_qa_pairs": len(all_pairs),
"evaluated_pairs": len(evaluated_pairs),
"unevaluated_pairs": len(all_pairs) - len(evaluated_pairs),
"evaluated_conversations": len(evaluated_ids)
}
def save_annotation(self,
conversation_id: str,
question: str,
original_answer: str,
improved_answer: str,
ratings: Dict[str, int],
notes: str = "") -> Tuple[bool, str]:
"""
Save evaluation annotation
"""
try:
# Create annotation object
annotation = {
"conversation_id": conversation_id,
"timestamp": datetime.datetime.now().isoformat(),
"question": question,
"original_answer": original_answer,
"improved_answer": improved_answer,
"ratings": ratings,
"notes": notes
}
# Create filename with conversation_id
filename = f"{self.annotations_path}/annotation_{conversation_id}.json"
# Convert to JSON bytes
json_content = json.dumps(annotation, ensure_ascii=False, indent=2).encode('utf-8')
# Upload to dataset using bytes buffer
self.api.upload_file(
path_or_fileobj=io.BytesIO(json_content),
path_in_repo=filename,
repo_id=self.dataset_id,
repo_type="dataset"
)
# Reset annotations cache
self._annotations = None
return True, "Annotation saved successfully"
except Exception as e:
logger.error(f"Error saving annotation: {e}")
return False, f"Failed to save annotation: {str(e)}"
def get_annotations(self, force_reload=False) -> List[Dict[str, Any]]:
"""
Get all saved annotations from dataset
Args:
force_reload: If True, force reload from dataset
"""
# Return cached data if available and not forcing reload
if self._annotations is not None and not force_reload:
logger.debug("Returning cached annotations")
return self._annotations
try:
annotations = []
files = self.api.list_repo_files(self.dataset_id, repo_type="dataset")
for file in files:
if file.startswith(f"{self.annotations_path}/annotation_") and file.endswith(".json"):
try:
# Download and parse annotation file
content = self.api.hf_hub_download(
repo_id=self.dataset_id,
filename=file,
repo_type="dataset"
)
with open(content, 'r', encoding='utf-8') as f:
annotation = json.load(f)
annotations.append(annotation)
except Exception as e:
logger.error(f"Error loading annotation {file}: {e}")
# Sort by timestamp (newest first)
annotations.sort(key=lambda x: x.get("timestamp", ""), reverse=True)
# Cache the results
self._annotations = annotations
return annotations
except Exception as e:
logger.error(f"Error getting annotations: {e}")
return []
def get_annotation(self, conversation_id: str) -> Optional[Dict[str, Any]]:
"""
Get specific annotation by conversation ID
"""
try:
# First check if annotations are loaded
if self._annotations is not None:
for annotation in self._annotations:
if annotation.get("conversation_id") == conversation_id:
return annotation
# If not found in cache, try direct file access
filename = f"{self.annotations_path}/annotation_{conversation_id}.json"
try:
content = self.api.hf_hub_download(
repo_id=self.dataset_id,
filename=filename,
repo_type="dataset"
)
with open(content, 'r', encoding='utf-8') as f:
return json.load(f)
except Exception as e:
logger.error(f"Error loading annotation for {conversation_id}: {e}")
return None
except Exception as e:
logger.error(f"Error getting annotation: {e}")
return None
def export_training_data(self, output_file: str, min_rating: int = 4) -> Tuple[bool, str]:
"""
Export high-quality annotated data for fine-tuning
Args:
output_file: Path to output file
min_rating: Minimum average rating to include in training data
Returns:
(success, message)
"""
annotations = self.get_annotations()
if not annotations:
return False, "No annotations available for export"
try:
# Filter annotations by quality
high_quality_examples = []
for annotation in annotations:
ratings = annotation.get("ratings", {})
# Calculate average rating
if ratings:
avg_rating = sum(ratings.values()) / len(ratings)
# Include only high-quality examples
if avg_rating >= min_rating:
high_quality_examples.append({
"messages": [
{"role": "user", "content": annotation.get("question", "")},
{"role": "assistant", "content": annotation.get("improved_answer", "")}
]
})
if not high_quality_examples:
return False, f"No examples meet the minimum quality threshold of {min_rating}"
# Save to JSONL format
with open(output_file, "w", encoding="utf-8") as f:
for example in high_quality_examples:
f.write(json.dumps(example, ensure_ascii=False) + "\n")
return True, f"Successfully exported {len(high_quality_examples)} high-quality examples for training"
except Exception as e:
return False, f"Error exporting training data: {str(e)}"
def generate_evaluation_report(self) -> Dict[str, Any]:
"""
Generate evaluation summary report
Returns:
Dictionary with evaluation metrics
"""
annotations = self.get_annotations()
if not annotations:
return {
"total_evaluations": 0,
"message": "No evaluations available"
}
# Initialize metrics
criteria = set()
for annotation in annotations:
criteria.update(annotation.get("ratings", {}).keys())
metrics = {
"total_evaluations": len(annotations),
"criteria_averages": {},
"overall_average": 0,
"improvement_rate": 0 # Percentage of answers that were improved
}
# Calculate averages for each criterion
for criterion in criteria:
values = [a.get("ratings", {}).get(criterion, 0) for a in annotations if criterion in a.get("ratings", {})]
if values:
metrics["criteria_averages"][criterion] = sum(values) / len(values)
# Calculate overall average
all_ratings = []
for annotation in annotations:
all_ratings.extend(annotation.get("ratings", {}).values())
if all_ratings:
metrics["overall_average"] = sum(all_ratings) / len(all_ratings)
# Calculate improvement rate
improved_count = sum(1 for a in annotations if a.get("original_answer") != a.get("improved_answer"))
metrics["improvement_rate"] = (improved_count / len(annotations)) * 100
return metrics
|