File size: 17,059 Bytes
19fc84f | 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 | import os
import json
import random
import logging
import numpy as np
from typing import Union, Optional
from tqdm import tqdm
from jsonschema import validate
from src.subclaim_processor.query_processor import IQueryProcessor
from src.common.llm.openai_rag_agent import OpenAIRAGAgent
from src.common.llm.openai_atomicfact_generator import OpenAIAtomicFactGenerator
from src.common.llm.openai_claim_verification import OpenAIClaimVerification
from src.subclaim_processor.scorer.base_scorer import IScorer
from src.subclaim_processor.scorer.document_scorer import IDocumentScorer
from src.calibration.utils import load_subclaim_data
class SubclaimProcessor(IQueryProcessor):
def __init__(
self,
faiss_manager,
response_model: str,
fact_generation_model: str,
claim_verification_model: str,
scorer: IScorer,
subclaims_file: str,
):
self.faiss_manager = faiss_manager
self.response_agent = OpenAIRAGAgent(faiss_manager, model=response_model)
self.generator = OpenAIAtomicFactGenerator(model=fact_generation_model)
self.verifier = OpenAIClaimVerification(model=claim_verification_model)
print(f"claim_verification_model: {claim_verification_model}")
self.scorer = scorer
self.subclaims_file = subclaims_file
with open(
"data/out/subclaims_schema.json", "r", encoding="utf-8"
) as schemafile:
self.subclaim_schema = json.load(schemafile)
def generate_responses(
self,
query_file: str,
top_k: int,
threshold: float,
response_temperature: float = 0.7,
truncation_strategy: Optional[Union[str, bool]] = "fixed_length",
truncate_by: Optional[str] = "\n",
):
"""Generate responses for queries"""
# Read queries
with open(query_file, "r", encoding="utf-8") as jsonfile:
queries = json.load(jsonfile)
responses = []
for query in tqdm(queries, desc="Generating responses"):
question = query["input"]
groups = query.get("groups", [])
# Document retrieval
retrieved_docs = self.faiss_manager.search_faiss_index(
question,
top_k=top_k,
threshold=threshold,
truncation_strategy=truncation_strategy,
truncate_by=truncate_by,
)
# Generate response
chat_response = self.response_agent.answer(
question, retrieved_docs, temperature=response_temperature, n_samples=1
)
response = chat_response.choices[0].message.content
responses.append(
{
"query": question,
"gld_ans": query["output"]["answer"],
"retrieved_docs": retrieved_docs,
"response": response,
"groups": groups,
"subclaims": [],
}
)
# Save responses
with open(self.subclaims_file, "w", encoding="utf-8") as f:
json.dump(responses, f, indent=4)
print(f"Responses saved to {self.subclaims_file}")
def get_subclaims_from_responses(self):
"""Process existing responses to extract subclaims and save updates incrementally"""
# Read existing responses
with open(self.subclaims_file, "r", encoding="utf-8") as jsonfile:
queries = json.load(jsonfile)
# Process each query and save updates in batches
batch_size = 10
for i in tqdm(range(0, len(queries), batch_size), desc="Extracting subclaims"):
batch = queries[i : i + batch_size]
modified = False
# Process each query in the batch
for query in batch:
if query["response"] and not query.get(
"subclaims"
): # Only process if no subclaims exist
try:
subclaims_with_log_probs = self.generator.get_facts_from_text(
query["response"]
)
query["subclaims"] = [
{
"subclaim": subclaim[0],
"scores": {
"log_prob": [score for token, score in subclaim[1]]
},
"annotations": {},
}
for subclaim in subclaims_with_log_probs
]
modified = True
except Exception as e:
print(
f"Error processing query: {query['query'][:50]}... Error: {str(e)}"
)
query["subclaims"] = (
[]
) # Add empty subclaims to mark as processed
modified = True
# Save updates if any changes were made in this batch
if modified:
with open(self.subclaims_file, "w", encoding="utf-8") as f:
json.dump(queries, f, indent=4)
print(
f"Saved updates through batch ending at index {min(i + batch_size, len(queries))}"
)
print(f"Completed subclaim extraction. Results saved in {self.subclaims_file}")
def score_subclaim(self, aggregation_strategy, scoring_strategy, lite: bool = False):
with open(self.subclaims_file, "r", encoding="utf-8") as jsonfile:
subclaims_data = json.load(jsonfile)
for entry in tqdm(subclaims_data, desc="Scoring subclaims"):
validate(instance=entry, schema=self.subclaim_schema)
for i, subclaim in enumerate(entry["subclaims"]):
if isinstance(self.scorer, IDocumentScorer):
if "noise" not in subclaim["scores"].keys():
subclaim["scores"]["noise"] = np.random.normal(0, 0.001)
if "relavance" not in subclaim["scores"].keys():
relavance_score = self.scorer.score(
claim=subclaim["subclaim"],
retrieved_docs=entry["retrieved_docs"],
aggregation_strategy=aggregation_strategy,
scoring_strategy=scoring_strategy,
)
subclaim["scores"]["relavance"] = float(relavance_score)
if (
"query_claim_cosine_similarity"
not in subclaim["scores"].keys()
):
query_claim_cosine_similarity = (
self.scorer.cosine_similarity(
subclaim["subclaim"], entry["query"]
)
)
subclaim["scores"]["query_claim_cosine_similarity"] = float(
query_claim_cosine_similarity
)
if not lite:
if (
"doc_claim_cosine_similarity"
not in subclaim["scores"].keys()
):
doc_claim_cosine_similarities = []
for doc in entry["retrieved_docs"]: # TODO
doc_claim_cosine_similarities.append(
self.scorer.cosine_similarity(
subclaim["subclaim"], doc
)
)
subclaim["scores"]["doc_claim_cosine_similarity"] = (
float(max(doc_claim_cosine_similarities))
if doc_claim_cosine_similarities
else 0
)
if not lite:
if "frequency" not in subclaim["scores"].keys():
frequency_score = self.scorer.frequency_score(
response_agent=self.response_agent,
question=entry["query"],
subclaim=subclaim["subclaim"],
retrived_docs=entry["retrieved_docs"],
temperature=1,
n_samples=5,
)
subclaim["scores"]["frequency"] = float(frequency_score)
if "random" not in subclaim["scores"].keys():
subclaim["scores"]["random"] = random.random()
if "ordinal" not in subclaim["scores"].keys():
subclaim["scores"]["ordinal"] = (
(i / len(entry["subclaims"]))
if len(entry["subclaims"]) > 0
else 0
)
if (
"min_log_prob" not in subclaim["scores"].keys()
and "log_prob" in subclaim["scores"].keys()
):
subclaim["scores"]["min_log_prob"] = min(
subclaim["scores"]["log_prob"]
)
with open(self.subclaims_file, "w", encoding="utf-8") as jsonfile:
json.dump(subclaims_data, jsonfile, indent=4)
print(f"Subclaims with scores saved to {self.subclaims_file}.")
def annotate_subclaim(self):
with open(self.subclaims_file, "r", encoding="utf-8") as jsonfile:
subclaims_data = json.load(jsonfile)
batch_size = 10
modified = False
for i in tqdm(
range(0, len(subclaims_data), batch_size),
desc="Annotating subclaims in batches",
):
batch = subclaims_data[i : i + batch_size]
for entry in batch:
try:
validate(instance=entry, schema=self.subclaim_schema)
# Skip if already annotated
if all(
subclaim.get("annotations", {}).get("gpt")
for subclaim in entry["subclaims"]
):
continue
doc_contents = []
for doc in entry["retrieved_docs"]:
try:
# Split the document string into page_content and metadata
doc_parts = doc.split("metadata=")
page_content = (
doc_parts[0].replace("page_content=", "").strip()
)
doc_contents.append(page_content)
except Exception as e:
doc_contents.append(f"Error processing document: {e}")
# Combine the formatted documents into a single context
context = "\n".join(doc_contents)
for subclaim in entry["subclaims"]:
if not subclaim.get("annotations", {}).get(
"gpt"
): # Only annotate if not already done
gold_answer = (
" ".join(entry["gld_ans"])
if isinstance(entry["gld_ans"], list)
else entry["gld_ans"]
)
annotation = self.verifier.annotate(
entry["query"],
gold_answer,
context,
subclaim["subclaim"],
)
if "annotations" not in subclaim:
subclaim["annotations"] = {}
subclaim["annotations"]["gpt"] = annotation
modified = True
except Exception as e:
logging.error(f"Error processing entry: {str(e)}")
continue
# Save after each batch if there were modifications
if modified:
try:
with open(self.subclaims_file, "w", encoding="utf-8") as jsonfile:
json.dump(subclaims_data, jsonfile, indent=4)
logging.info(
f"Saved batch through index {min(i + batch_size, len(subclaims_data))}"
)
modified = False # Reset modified flag
except Exception as e:
logging.error(f"Error saving batch: {str(e)}")
logging.info(f"Completed annotation. Results saved in {self.subclaims_file}")
def process_subclaims(
query_path,
subclaims_path,
faiss_manager,
scorer,
config,
lite: bool = False,
):
truncation_strategy = config["index"]["truncation_config"]["strategy"]
truncate_by = config["index"]["truncation_config"]["truncate_by"]
top_k = config["rag"]["retrival_topk"]
threshold = config["rag"]["retrival_threshold"]
response_model = config["rag"]["response_model"]
response_temperature = config["rag"]["response_temperature"]
fact_generation_model = config["rag"]["fact_generation_model"]
aggregation_strategy = config["conformal_prediction"]["aggregation_strategy"]
scoring_strategy = config["conformal_prediction"]["scoring_strategy"]
claim_verification_model = config["conformal_prediction"][
"claim_verification_model"
]
# Check if the file exists and load it if it does
data = None
if os.path.exists(subclaims_path):
data = load_subclaim_data(subclaims_path)
# Check if the data is valid
score_method_to_check = [
"noise",
"relavance",
"query_claim_cosine_similarity",
"random",
"ordinal",
"min_log_prob",
]
if not lite:
score_method_to_check += ["frequency", "doc_claim_cosine_similarity"]
if data:
needs_subclaim = any(len(pt["subclaims"]) == 0 for pt in data)
if needs_subclaim:
needs_scoring = True
needs_annotation = True
else:
needs_scoring = any(
len(subclaim["scores"]) == 0
for pt in data
for subclaim in pt["subclaims"]
) or any(
score_method not in subclaim["scores"].keys()
for pt in data
for subclaim in pt["subclaims"]
for score_method in score_method_to_check
)
needs_annotation = any(
len(subclaim["annotations"]) == 0
for pt in data
for subclaim in pt["subclaims"]
)
if not (needs_subclaim or needs_scoring or needs_annotation):
print(f"Subclaims data already exists in {subclaims_path}.")
return data
# Initialize processor only when needed
processor = SubclaimProcessor(
faiss_manager,
response_model,
fact_generation_model,
claim_verification_model,
scorer,
subclaims_path,
)
# Generate subclaims if data doesn't exist
if not data:
processor.generate_responses(
query_path,
top_k=top_k,
threshold=threshold,
response_temperature=response_temperature,
truncation_strategy=truncation_strategy,
truncate_by=truncate_by,
)
processor.get_subclaims_from_responses()
processor.score_subclaim(
aggregation_strategy=aggregation_strategy,
scoring_strategy=scoring_strategy,
lite=lite,
)
processor.annotate_subclaim()
else:
# Update only what's needed
if needs_subclaim:
processor.get_subclaims_from_responses()
if needs_scoring:
processor.score_subclaim(
aggregation_strategy=aggregation_strategy,
scoring_strategy=scoring_strategy,
lite=lite,
)
if needs_annotation:
processor.annotate_subclaim()
return load_subclaim_data(subclaims_path)
|