Spaces:
Paused
Paused
File size: 21,456 Bytes
45ee481 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 | """
Q&A Generator Module
Generate synthetic Q&A training pairs from blog content using Claude/GPT-4 API.
Creates diverse questions and CEO-style answers for fine-tuning.
Example usage:
generator = QAGenerator(provider="anthropic")
qa_pairs = generator.generate_from_segments(segments, num_pairs=500)
"""
import json
import os
import random
import time
from dataclasses import dataclass, field
from pathlib import Path
from typing import Literal, Optional
from loguru import logger
from tenacity import retry, stop_after_attempt, wait_exponential
try:
import anthropic
ANTHROPIC_AVAILABLE = True
except ImportError:
ANTHROPIC_AVAILABLE = False
try:
import openai
OPENAI_AVAILABLE = True
except ImportError:
OPENAI_AVAILABLE = False
@dataclass
class QAPair:
"""Represents a Q&A training pair."""
question: str
answer: str
source_segment_index: int
source_post_title: str
question_type: str
metadata: dict = field(default_factory=dict)
def to_dict(self) -> dict:
"""Convert to dictionary for serialization."""
return {
"question": self.question,
"answer": self.answer,
"source_segment_index": self.source_segment_index,
"source_post_title": self.source_post_title,
"question_type": self.question_type,
"metadata": self.metadata,
}
# Question type templates for diverse generation
QUESTION_TEMPLATES = {
"opinion": [
"What is your view on {topic}?",
"How do you feel about {topic}?",
"What's your take on {topic}?",
"Do you think {topic} is important? Why?",
"What are your thoughts on {topic}?",
],
"strategic": [
"How should companies approach {topic}?",
"What strategy would you recommend for {topic}?",
"What's the best way to handle {topic}?",
"How do you see {topic} evolving in the future?",
"What opportunities do you see in {topic}?",
],
"personal_philosophy": [
"What drives your passion for {topic}?",
"What lessons have you learned about {topic}?",
"How has your thinking on {topic} evolved?",
"What advice would you give about {topic}?",
"What's the most important thing to understand about {topic}?",
],
"factual": [
"Can you explain {topic}?",
"What is {topic} and why does it matter?",
"Tell me about your experience with {topic}.",
"What are the key aspects of {topic}?",
"How does {topic} work in practice?",
],
"challenge": [
"Some people criticize {topic}. How would you respond?",
"What are the main challenges with {topic}?",
"What mistakes do people commonly make regarding {topic}?",
"Is there a downside to {topic}?",
"What are the risks associated with {topic}?",
],
}
class QAGenerator:
"""
Generate synthetic Q&A pairs using LLM APIs.
Supports:
- Anthropic Claude API
- OpenAI GPT-4 API
- Rate limiting and retry logic
- Cost estimation
Example:
>>> generator = QAGenerator(provider="anthropic")
>>> pairs = generator.generate_from_segments(segments, num_pairs=100)
>>> print(f"Generated {len(pairs)} Q&A pairs")
"""
# Pricing per 1M tokens (approximate, check current rates)
PRICING = {
"anthropic": {"input": 3.0, "output": 15.0}, # Claude 3 Sonnet
"openai": {"input": 10.0, "output": 30.0}, # GPT-4
}
def __init__(
self,
provider: Literal["anthropic", "openai"] = "anthropic",
model: Optional[str] = None,
api_key: Optional[str] = None,
requests_per_minute: int = 20,
ceo_name: str = "Ryouken Okuni",
company_name: str = "Akatsuki AI Technologies",
):
"""
Initialize the Q&A generator.
Args:
provider: API provider ("anthropic" or "openai")
model: Model name (defaults based on provider)
api_key: API key (or uses environment variable)
requests_per_minute: Rate limit
ceo_name: Name of the CEO persona
company_name: Name of the company
"""
self.provider = provider
self.requests_per_minute = requests_per_minute
self.ceo_name = ceo_name
self.company_name = company_name
# Set default models
if model is None:
self.model = (
"claude-sonnet-4-20250514" if provider == "anthropic"
else "gpt-4-turbo-preview"
)
else:
self.model = model
# Initialize client
if provider == "anthropic":
if not ANTHROPIC_AVAILABLE:
raise ImportError("anthropic package not installed. Run: pip install anthropic")
api_key = api_key or os.environ.get("ANTHROPIC_API_KEY")
if not api_key:
raise ValueError("ANTHROPIC_API_KEY not found in environment")
self.client = anthropic.Anthropic(api_key=api_key)
else:
if not OPENAI_AVAILABLE:
raise ImportError("openai package not installed. Run: pip install openai")
api_key = api_key or os.environ.get("OPENAI_API_KEY")
if not api_key:
raise ValueError("OPENAI_API_KEY not found in environment")
self.client = openai.OpenAI(api_key=api_key)
# Rate limiting
self._last_request_time = 0
self._min_request_interval = 60.0 / requests_per_minute
# Token tracking for cost estimation
self._total_input_tokens = 0
self._total_output_tokens = 0
def estimate_cost(self, num_pairs: int, avg_segment_tokens: int = 400) -> dict:
"""
Estimate API cost before generation.
Args:
num_pairs: Number of Q&A pairs to generate
avg_segment_tokens: Average tokens per segment
Returns:
Dictionary with estimated costs
"""
# Estimate tokens per pair
prompt_tokens = 500 + avg_segment_tokens # System + segment
completion_tokens = 300 # Average response
total_input = num_pairs * prompt_tokens
total_output = num_pairs * completion_tokens
pricing = self.PRICING[self.provider]
input_cost = (total_input / 1_000_000) * pricing["input"]
output_cost = (total_output / 1_000_000) * pricing["output"]
return {
"estimated_input_tokens": total_input,
"estimated_output_tokens": total_output,
"estimated_cost_usd": round(input_cost + output_cost, 2),
"provider": self.provider,
"model": self.model,
}
def get_actual_cost(self) -> dict:
"""Get actual cost based on tracked tokens."""
pricing = self.PRICING[self.provider]
input_cost = (self._total_input_tokens / 1_000_000) * pricing["input"]
output_cost = (self._total_output_tokens / 1_000_000) * pricing["output"]
return {
"total_input_tokens": self._total_input_tokens,
"total_output_tokens": self._total_output_tokens,
"actual_cost_usd": round(input_cost + output_cost, 2),
}
def generate_from_segments(
self,
segments: list,
num_pairs: int = 500,
questions_per_segment: int = 3,
output_path: Optional[str | Path] = None,
) -> list[QAPair]:
"""
Generate Q&A pairs from text segments.
Args:
segments: List of TextSegment objects
num_pairs: Target number of Q&A pairs
questions_per_segment: Max questions per segment
output_path: Optional path to save pairs as JSON
Returns:
List of QAPair objects
"""
logger.info(f"Generating {num_pairs} Q&A pairs from {len(segments)} segments")
# Estimate cost first
estimate = self.estimate_cost(num_pairs)
logger.info(f"Estimated cost: ${estimate['estimated_cost_usd']:.2f}")
qa_pairs = []
segments_to_use = list(segments)
random.shuffle(segments_to_use)
pairs_generated = 0
segment_idx = 0
while pairs_generated < num_pairs and segment_idx < len(segments_to_use):
segment = segments_to_use[segment_idx]
# Generate questions for this segment
questions_for_segment = min(
questions_per_segment,
num_pairs - pairs_generated,
)
try:
segment_pairs = self._generate_for_segment(
segment, questions_for_segment
)
qa_pairs.extend(segment_pairs)
pairs_generated += len(segment_pairs)
logger.debug(
f"Generated {len(segment_pairs)} pairs from segment {segment_idx} "
f"({pairs_generated}/{num_pairs} total)"
)
except Exception as e:
logger.warning(f"Failed to generate for segment {segment_idx}: {e}")
segment_idx += 1
# Progress update
if pairs_generated % 50 == 0:
logger.info(f"Progress: {pairs_generated}/{num_pairs} pairs generated")
# Log final cost
actual_cost = self.get_actual_cost()
logger.info(f"Actual cost: ${actual_cost['actual_cost_usd']:.2f}")
# Save if path provided
if output_path:
self._save_pairs(qa_pairs, output_path)
return qa_pairs
@retry(
stop=stop_after_attempt(3),
wait=wait_exponential(multiplier=1, min=4, max=60),
)
def _generate_for_segment(
self, segment, num_questions: int
) -> list[QAPair]:
"""Generate Q&A pairs for a single segment."""
# Rate limiting
self._rate_limit()
# Select question types
question_types = random.sample(
list(QUESTION_TEMPLATES.keys()),
min(num_questions, len(QUESTION_TEMPLATES)),
)
# Build prompt
system_prompt = self._build_system_prompt()
user_prompt = self._build_generation_prompt(
segment.content, question_types, num_questions
)
# Call API
response_text, input_tokens, output_tokens = self._call_api(
system_prompt, user_prompt
)
# Track tokens
self._total_input_tokens += input_tokens
self._total_output_tokens += output_tokens
# Parse response
pairs = self._parse_response(
response_text,
segment.segment_index,
segment.source_post_title,
question_types,
)
return pairs
def _rate_limit(self) -> None:
"""Enforce rate limiting between requests."""
current_time = time.time()
time_since_last = current_time - self._last_request_time
if time_since_last < self._min_request_interval:
sleep_time = self._min_request_interval - time_since_last
time.sleep(sleep_time)
self._last_request_time = time.time()
def _build_system_prompt(self) -> str:
"""Build the system prompt for Q&A generation."""
return f"""You are helping create training data for an AI assistant that will replicate the communication style of {self.ceo_name}, CEO of {self.company_name}.
Your task is to generate realistic Q&A pairs based on provided blog content. The questions should be ones that stakeholders, employees, journalists, or business partners might ask. The answers should authentically capture the CEO's voice, reasoning patterns, and communication style as demonstrated in the source content.
Guidelines for generating responses:
1. Match the tone and vocabulary of the original content
2. Preserve the CEO's unique way of explaining concepts
3. Maintain the same level of formality/informality
4. Include similar rhetorical patterns (questions, examples, analogies)
5. Stay factually consistent with the source material
6. For topics not directly covered, extrapolate based on evident principles and values
Output format: Return a JSON array of Q&A pairs. Each pair should have:
- "question": The stakeholder's question
- "answer": The CEO's response in their authentic voice
- "question_type": The category of question (opinion, strategic, factual, etc.)"""
def _build_generation_prompt(
self, content: str, question_types: list[str], num_questions: int
) -> str:
"""Build the user prompt for generation."""
types_str = ", ".join(question_types)
return f"""Based on the following blog content, generate {num_questions} Q&A pairs.
Include these question types: {types_str}
Blog content:
---
{content}
---
Generate {num_questions} diverse Q&A pairs that capture the CEO's authentic voice. Return only valid JSON array.
Example format:
[
{{
"question": "What is your view on AI in business?",
"answer": "I believe AI is fundamentally transforming how we...",
"question_type": "opinion"
}}
]"""
def _call_api(
self, system_prompt: str, user_prompt: str
) -> tuple[str, int, int]:
"""Call the LLM API and return response with token counts."""
if self.provider == "anthropic":
response = self.client.messages.create(
model=self.model,
max_tokens=2000,
system=system_prompt,
messages=[{"role": "user", "content": user_prompt}],
)
text = response.content[0].text
input_tokens = response.usage.input_tokens
output_tokens = response.usage.output_tokens
else:
response = self.client.chat.completions.create(
model=self.model,
max_tokens=2000,
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt},
],
)
text = response.choices[0].message.content
input_tokens = response.usage.prompt_tokens
output_tokens = response.usage.completion_tokens
return text, input_tokens, output_tokens
def _parse_response(
self,
response_text: str,
segment_index: int,
source_title: str,
question_types: list[str],
) -> list[QAPair]:
"""Parse the API response into QAPair objects."""
pairs = []
try:
# Try to extract JSON from response
# Handle markdown code blocks
if "```json" in response_text:
json_start = response_text.find("```json") + 7
json_end = response_text.find("```", json_start)
response_text = response_text[json_start:json_end]
elif "```" in response_text:
json_start = response_text.find("```") + 3
json_end = response_text.find("```", json_start)
response_text = response_text[json_start:json_end]
data = json.loads(response_text.strip())
if isinstance(data, list):
for item in data:
if isinstance(item, dict) and "question" in item and "answer" in item:
pairs.append(QAPair(
question=item["question"],
answer=item["answer"],
source_segment_index=segment_index,
source_post_title=source_title,
question_type=item.get("question_type", "unknown"),
))
except json.JSONDecodeError as e:
logger.warning(f"Failed to parse JSON response: {e}")
# Try to salvage partial response
pairs = self._salvage_partial_response(
response_text, segment_index, source_title
)
return pairs
def _salvage_partial_response(
self, response_text: str, segment_index: int, source_title: str
) -> list[QAPair]:
"""Attempt to extract Q&A pairs from malformed response."""
pairs = []
# Look for question/answer patterns
import re
qa_pattern = re.compile(
r'"question":\s*"([^"]+)".*?"answer":\s*"([^"]+)"',
re.DOTALL
)
for match in qa_pattern.finditer(response_text):
pairs.append(QAPair(
question=match.group(1),
answer=match.group(2),
source_segment_index=segment_index,
source_post_title=source_title,
question_type="unknown",
))
return pairs
def _save_pairs(self, pairs: list[QAPair], path: str | Path) -> None:
"""Save Q&A pairs to JSON file."""
path = Path(path)
data = [p.to_dict() for p in pairs]
with open(path, "w", encoding="utf-8") as f:
json.dump(data, f, indent=2, ensure_ascii=False)
logger.info(f"Saved {len(pairs)} Q&A pairs to: {path}")
@staticmethod
def load_pairs(path: str | Path) -> list[QAPair]:
"""Load Q&A pairs from JSON file."""
with open(path, "r", encoding="utf-8") as f:
data = json.load(f)
return [
QAPair(
question=item["question"],
answer=item["answer"],
source_segment_index=item["source_segment_index"],
source_post_title=item["source_post_title"],
question_type=item["question_type"],
metadata=item.get("metadata", {}),
)
for item in data
]
def main():
"""CLI entry point for testing the generator."""
import argparse
parser = argparse.ArgumentParser(
description="Generate Q&A pairs from text segments using LLM APIs",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
python qa_generator.py segments.json --output qa_pairs.json --num-pairs 100
python qa_generator.py segments.json --provider openai --estimate-only
Environment variables:
ANTHROPIC_API_KEY - Anthropic API key (for Claude)
OPENAI_API_KEY - OpenAI API key (for GPT-4)
""",
)
parser.add_argument("input", help="Input segments JSON file")
parser.add_argument("--output", "-o", help="Output Q&A pairs JSON file")
parser.add_argument(
"--num-pairs",
type=int,
default=100,
help="Number of Q&A pairs to generate (default: 100)",
)
parser.add_argument(
"--provider",
choices=["anthropic", "openai"],
default="anthropic",
help="API provider (default: anthropic)",
)
parser.add_argument(
"--model",
help="Model name (defaults based on provider)",
)
parser.add_argument(
"--estimate-only",
action="store_true",
help="Only show cost estimate, don't generate",
)
parser.add_argument(
"--ceo-name",
default="Ryouken Okuni",
help="CEO name for persona",
)
parser.add_argument(
"--company-name",
default="Akatsuki AI Technologies",
help="Company name for persona",
)
args = parser.parse_args()
# Load segments
with open(args.input, "r", encoding="utf-8") as f:
segments_data = json.load(f)
# Convert to simple objects for the generator
from dataclasses import dataclass as dc
@dc
class SimpleSegment:
content: str
segment_index: int
source_post_title: str
segments = [
SimpleSegment(
content=s["content"],
segment_index=s.get("segment_index", i),
source_post_title=s.get("source_post_title", "Unknown"),
)
for i, s in enumerate(segments_data)
]
try:
generator = QAGenerator(
provider=args.provider,
model=args.model,
ceo_name=args.ceo_name,
company_name=args.company_name,
)
except (ImportError, ValueError) as e:
print(f"Error initializing generator: {e}")
return 1
# Show estimate
estimate = generator.estimate_cost(args.num_pairs)
print(f"\n=== Cost Estimate ===")
print(f"Provider: {estimate['provider']}")
print(f"Model: {estimate['model']}")
print(f"Estimated input tokens: {estimate['estimated_input_tokens']:,}")
print(f"Estimated output tokens: {estimate['estimated_output_tokens']:,}")
print(f"Estimated cost: ${estimate['estimated_cost_usd']:.2f}")
if args.estimate_only:
return 0
# Confirm
print("\nProceed with generation? [y/N] ", end="")
response = input().strip().lower()
if response != "y":
print("Cancelled.")
return 0
# Generate
output_path = args.output or "qa_pairs.json"
pairs = generator.generate_from_segments(
segments, num_pairs=args.num_pairs, output_path=output_path
)
# Show results
actual = generator.get_actual_cost()
print(f"\n=== Generation Complete ===")
print(f"Generated: {len(pairs)} Q&A pairs")
print(f"Actual cost: ${actual['actual_cost_usd']:.2f}")
print(f"Saved to: {output_path}")
return 0
if __name__ == "__main__":
exit(main())
|