Spaces:
Sleeping
Sleeping
File size: 36,039 Bytes
bb1e5d5 d6b0e28 bb1e5d5 5351c4a bb1e5d5 7dadf03 bb1e5d5 49f8037 bb1e5d5 35facc7 bb1e5d5 a505e26 8ac474e bb1e5d5 8ac474e bb1e5d5 a505e26 bb1e5d5 35facc7 bb1e5d5 35facc7 bb1e5d5 35facc7 a505e26 bb1e5d5 fa76780 bb1e5d5 a505e26 35facc7 bb1e5d5 35facc7 bb1e5d5 35facc7 bb1e5d5 35facc7 bb1e5d5 241b4ce bb1e5d5 241b4ce bb1e5d5 a505e26 bb1e5d5 35facc7 bb1e5d5 38067cb bb1e5d5 17e05cb bb1e5d5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 |
import os
import time
import asyncio
import hashlib
from typing import Optional
from datetime import datetime
import re
from concurrent.futures import ThreadPoolExecutor
from typing import List, Dict, Any
from groq import Groq
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline, AutoModelForCausalLM
import torch
import numpy as np
from core.models.knowledge_base import OptimizedGazaKnowledgeBase
from core.fact_checker import MedicalFactChecker, clean_ocr_artifacts
from core.utils.config import (
MEDICAL_SYSTEM_PROMPT,
GROQ_API_KEY,
FLAN_MODEL_NAME,
FALLBACK_MODEL_NAME,
MAX_CACHE_SIZE,
MAX_CONTEXT_CHARS
)
from transformers import pipeline
print("🧪 Logger test: ", 'logger' in globals())
from core.utils.logger import logger
from transformers import pipeline
class ArabicTranslator:
def __init__(self):
try:
self.en_to_ar = pipeline("translation", model="facebook/m2m100_418M", src_lang="en", tgt_lang="ar")
self.ar_to_en = pipeline("translation", model="facebook/m2m100_418M", src_lang="ar", tgt_lang="en")
print("✅ Translation models loaded")
except Exception as e:
print(f"❌ Failed to load translation models: {e}")
self.en_to_ar = None
self.ar_to_en = None
def translate_to_english(self, text):
if not self.ar_to_en:
print("⚠️ Arabic-to-English translation model not available.")
return text
return self.ar_to_en(text[:1000], max_length=1024)[0]["translation_text"]
def translate_to_arabic(self, text):
if not self.en_to_ar:
print("⚠️ English-to-Arabic translation model not available.")
return text
# Strip markdown artifacts (prevent "* * *")
clean_text = re.sub(r'[*_`~#>]', '', text)
MAX_INPUT = 900 # Stay below token limits
if len(clean_text) > MAX_INPUT:
print(f"⚠️ Input too long ({len(clean_text)}), truncating for translation.")
clean_text = clean_text[:MAX_INPUT]
return self.en_to_ar(clean_text, max_length=1024)[0]["translation_text"]
def translate(self, text: str, direction: str = "to_en") -> str:
if direction == "to_en":
return self.translate_to_english(text)
elif direction == "to_ar":
return self.translate_to_arabic(text)
else:
raise ValueError("Invalid translation direction: choose 'to_en' or 'to_ar'")
class OptimizedGazaRAGSystem:
"""Optimized RAG system using pre-made assets"""
def __init__(self, vector_store_dir: str = "./vector_store"):
self.knowledge_base = OptimizedGazaKnowledgeBase(vector_store_dir)
self.fact_checker = MedicalFactChecker()
self.groq_client = None
self.llm = None
self.tokenizer = None
self.use_native_generation = True # or False by config/env
self.system_prompt = self._create_system_prompt()
self.arabic_translator = ArabicTranslator()
self.generation_pipeline = None
self.response_cache = {}
self.executor = ThreadPoolExecutor(max_workers=2)
self.definitive_patterns = [
re.compile(r, re.IGNORECASE) for r in [
r'will\s+(?:cure|heal|fix)\b', # Only block definitive claims
r'guaranteed\s+to',
r'completely\s+(?:safe|effective)\b',
r'\b(?:inject|syringe)\b' # Added dangerous procedures
]
]
translated_test = self.arabic_translator.translate("How do I treat a wound?")
print("🔥 Arabic test translation:", translated_test)
def initialize(self):
"""Initialize the optimized RAG system"""
logger.info("🚀 Initializing Optimized Gaza RAG System...")
self.knowledge_base.initialize()
logger.info("✅ Optimized Gaza RAG System ready!")
def _initialize_groq(self):
"""Initialize Groq client with proper error handling"""
try:
api_key = os.getenv("GROQ_API_KEY")
if not api_key:
logger.warning("⚠️ GROQ_API_KEY environment variable not set")
return None
client = Groq(api_key=api_key)
# Test the connection with a simple API call
try:
client.models.list() # Simple API call to verify key
logger.info("✅ Groq client initialized successfully")
return client
except Exception as auth_error:
logger.error(f"❌ Groq API key invalid: {auth_error}")
return None
except Exception as e:
logger.error(f"❌ Groq initialization failed: {e}")
return None
def generate_raw_text(self, prompt: str) -> str:
"""Direct text generation without RAG, safety checks, or translation."""
if not self.generation_pipeline:
self._initialize_llm()
output = self.generation_pipeline(prompt)
return output[0]["generated_text"].strip() if output else ""
def _format_kb_response(self, text: str) -> str:
"""Conditionally expands short KB entries using FLAN→Groq pipeline"""
clean_text = clean_ocr_artifacts(text).strip()
def _get_groq_client():
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
# FIXED HEURISTIC: Increased word count threshold and narrowed keyword list
is_detailed = len(clean_text.split()) > 200 and any(
kw in clean_text.lower()
for kw in ["fracture", "bleeding", "wound", "infection"]
)
if is_detailed:
return f"📚 **Comprehensive Medical Guidance:**\n\n{clean_text}"
# Otherwise, enrich it dynamically using FLAN + Groq
try:
refined_prompt = self._create_prompt_from_rag(query=clean_text, rag_results=[])
enriched = self._generate_with_groq(query=clean_text, refined_prompt=refined_prompt)
return f"📚 **Comprehensive Medical Guidance:**\n\n{enriched}"
except Exception as e:
logger.warning(f"[FormatKB] Enrichment failed: {e}")
return f"📚 **Basic Medical Information:**\n\n{clean_text}\n\n⚠️ Could not expand this content automatically."
def _create_system_prompt(self) -> str:
"""Enhanced system prompt for Gaza context"""
MEDICAL_SYSTEM_PROMPT = """
[STRICT GAZA MEDICAL PROTOCOL]
You are a WHO-certified medical assistant for Gaza. You MUST:
1. Follow WHO war-zone protocols
2. Reject unsafe treatments (ESPECIALLY syringe use for burns)
3. Prioritize resource-scarce solutions
4. Add Islamic medical considerations
5. Format responses clearly with:
- 🩹 Immediate Actions
- ⚠️ Contraindications
- 💡 Resource Alternatives
6. Include source references [Source X]
7. Always add: "📞 Verify with Gaza Red Crescent (101)" for serious cases
OUTPUT EXAMPLE:
### Burn Treatment ###
🩹 Cool with clean water for 10-20 mins [Source 1]
⚠️ Never apply ice directly [Source 2]
💡 Use clean damp cloth if water scarce [Source 3]
📍 Gaza Context: Adapt based on available supplies
📞 Verify with Gaza Red Crescent (101) if severe
"""
return MEDICAL_SYSTEM_PROMPT
def _initialize_llm(self):
"""Load medical FLAN-T5 model with proper error handling and optimizations"""
model_name = "rivapereira123/medical-flan-t5"
try:
logger.info(f"🔄 Loading medical FLAN-T5 model: {model_name}")
self.tokenizer = AutoTokenizer.from_pretrained(
model_name,
cache_dir="./model_cache" # Optional local caching
)
self.llm = AutoModelForSeq2SeqLM.from_pretrained(
model_name,
low_cpu_mem_usage=True, # Critical for CPU
)
self.generation_pipeline = pipeline(
"text2text-generation",
model=self.llm,
tokenizer=self.tokenizer,
max_length=512, # Prevent OOM errors
truncation=True
)
logger.info("✅ Medical FLAN-T5 loaded successfully (CPU mode)")
except Exception as e:
logger.error(f"❌ Critical error loading model: {str(e)}")
logger.warning("⚠️ Medical QA features will be disabled")
self.llm = None
self.tokenizer = None
self.generation_pipeline = None
def _initialize_fallback_llm(self):
"""Enhanced fallback model with better error handling"""
try:
logger.info("🔄 Loading fallback model...")
fallback_model = "microsoft/DialoGPT-small"
self.tokenizer = AutoTokenizer.from_pretrained(fallback_model)
self.llm = AutoModelForCausalLM.from_pretrained(
fallback_model,
torch_dtype=torch.float32,
low_cpu_mem_usage=True
)
if self.tokenizer.pad_token is None:
self.tokenizer.pad_token = self.tokenizer.eos_token
self.generation_pipeline = pipeline(
"text-generation",
model=self.llm,
tokenizer=self.tokenizer,
return_full_text=False
)
logger.info("✅ Fallback model loaded successfully")
except Exception as e:
logger.error(f"❌ Fallback model failed: {e}")
self.llm = None
self.generation_pipeline = None
async def generate_response_async(self, query: str, progress_callback=None, language="English") -> Dict[str, Any]:
"""Async response generation with progress tracking"""
start_time = time.time()
# Step 0: Translate Arabic → English if needed
original_query = query
original_language = language
is_arabic_request = language.lower() == "arabic" or self._is_arabic(query)
if is_arabic_request and self.arabic_translator:
logger.info("🈸 Arabic input detected - translating to English for processing")
query = self.arabic_translator.translate_to_english(query)
if progress_callback:
progress_callback(0.1, "🔍 Checking cache...")
# Check cache first (using English query for cache key)
# Check cache first (using English query for cache key)
query_hash = hashlib.md5(query.encode()).hexdigest()
if query_hash in self.response_cache:
cached_response = self.response_cache[query_hash]
# Translate cached response if needed
if is_arabic_request and self.arabic_translator:
cached_response["response"] = self.arabic_translator.translate_to_arabic(cached_response["response"])
cached_response["cached"] = True
cached_response["response_time"] = 0.1
if progress_callback:
progress_callback(1.0, "💾 Retrieved from cache!")
return cached_response
try:
if progress_callback:
progress_callback(0.2, "🤖 Initializing LLM...")
# Initialize LLM only when needed
if self.llm is None:
await asyncio.get_event_loop().run_in_executor(
self.executor, self._initialize_llm
)
if progress_callback:
progress_callback(0.4, "🔍 Searching knowledge base...")
# Enhanced knowledge retrieval using pre-made index
search_results = await asyncio.get_event_loop().run_in_executor(
self.executor, self.knowledge_base.search, query, 5
)
if progress_callback:
progress_callback(0.6, "📝 Preparing context...")
context = self._prepare_context(search_results)
if progress_callback:
progress_callback(0.8, "🧠 Generating response...")
# Generate response
english_response = await asyncio.get_event_loop().run_in_executor(
self.executor, self._generate_response, query, context
)
if progress_callback:
progress_callback(0.9, "🛡️ Validating safety...")
# Enhanced safety check
safety_check = self.fact_checker.check_medical_accuracy(english_response, context)
# Step 2: Translate response → Arabic if needed
# Prepare final response structure
final_response = self._prepare_final_response(
english_response,
search_results,
safety_check,
time.time() - start_time
)
# Step 3: Translate final response to Arabic if requested
if is_arabic_request and self.arabic_translator:
logger.info("🌐 Translating final response to Arabic")
final_response["response"] = self.arabic_translator.translate_to_arabic(final_response["response"])
final_response["translated"] = True
final_response["original_language"] = "Arabic"
else:
final_response["translated"] = False
final_response["original_language"] = "English"
# Cache the English version (for consistency)
if len(self.response_cache) < 100:
english_cache_response = final_response.copy()
english_cache_response["response"] = english_response # Store English version
self.response_cache[query_hash] = english_cache_response
if progress_callback:
progress_callback(1.0, "✅ Complete!")
return final_response
except Exception as e:
logger.error(f"❌ Error generating response: {e}")
if progress_callback:
progress_callback(1.0, f"❌ Error: {str(e)}")
return self._create_error_response(str(e))
# def _generate_with_flan(self, query: str, context: Optional[str] = None) -> str:
# """Generate a response using the FLAN model directly (no Groq)."""
# if not self.generation_pipeline:
# raise RuntimeError("FLAN generation pipeline not initialized")
# # Build simple instructional prompt
# prompt = f"""
# You are a medical assistant working in Gaza.
# Query:
# {query}
# Context:
# {context if context else "No additional context"}
# Respond with:
# - 🩹 Immediate Actions
# - ⚠️ Contraindications
# - 💡 Alternatives
# - 🚨 When to seek emergency help
# """.strip()
# result = self.generation_pipeline(prompt)
# return result[0]["generated_text"].strip() if result else "⚠️ No response generated."
def _create_prompt_from_rag(self, query: str, rag_results: List[Dict[str, Any]]) -> str:
"""Use FLAN-T5 to condense RAG results into a clean prompt"""
if not self.llm or not rag_results:
return query # Fallback to original query
# Create context string from RAG results
context = "\n".join([f"[Source {i+1}]: {res['text']}"
for i, res in enumerate(rag_results[:6])])
# Create prompt for FLAN
prompt = f"""You are a medical researcher, Expand this medical context into a concise prompt for detailed response:
Original Query: {query}
Context:
{context}
Create a comprehensive response that includes:
1. Step-by-step treatment instructions
2. Gaza-specific adaptations
3. Alternative methods for resource-limited situations
4. Clear danger signs requiring professional care
5. Proper wound care timeline
Create a detailed question incorporating key points from the context, Structure your response with:
- Immediate Actions - Contraindications - Follow-up Care - Emergency Indicators:"""
# Generate with FLAN-T5
inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
outputs = self.llm.generate(
input_ids=inputs.input_ids,
max_new_tokens=200,
num_beams=3,
early_stopping=True
)
refined_prompt = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
return f"{refined_prompt}\n\nContext References:\n{context}"
def _is_arabic(self, text):
return any('\u0600' <= c <= '\u06FF' for c in text)
def _prepare_context(self, search_results: List[Dict[str, Any]]) -> str:
MAX_CHARS = 1500
if not search_results:
return "First aid protocol: "
context_parts = []
for result in search_results[:5]: # Top 3 results only
text = str(result.get('text', '')).strip()
context_parts.append({
'text': clean_ocr_artifacts(text),
'source': result.get('source', 'unknown'),
'score': result.get('score', 0.0)
})
return "\n\n".join(
f"[Reference {i+1}]: {ctx['text']}"
for i, ctx in enumerate(context_parts)
)[:MAX_CHARS]
def _format_context_with_groq(self, context_parts: List[Dict[str, Any]]) -> str:
"""Format context using Groq with comprehensive error handling"""
def _get_groq_client(self):
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
logger.info(f"GROQ_API_KEY prefix-format context with groq: {os.getenv('GROQ_API_KEY')[:5]}****")
if not hasattr(self, 'groq_client') or not self.groq_client:
raise ValueError("Groq client not initialized")
if not context_parts:
return "No context available"
try:
# Prepare context string
context_str = "\n\n".join(
f"Source {i+1} ({ctx['source']}, relevance {ctx['score']:.2f}):\n{ctx['text']}"
for i, ctx in enumerate(context_parts)
)
response = self.groq_client.chat.completions.create(
model="deepseek-r1-distill-llama-70b",
messages=[
{
"role": "system",
"content": self.system_prompt
},
{
"role": "user",
"content": f"Organize this medical context:\n\n{context_str}"
}
],
temperature=0.3,
max_tokens=2000,
top_p=0.9
)
# Validate response structure
if not response or not response.choices:
raise ValueError("Empty Groq response")
if not hasattr(response.choices[0], 'message') or not hasattr(response.choices[0].message, 'content'):
raise ValueError("Invalid response format")
formatted = response.choices[0].message.content
if formatted is None:
raise ValueError("Empty response content")
# Post-processing safety checks
if not isinstance(formatted, str):
raise ValueError("Formatted context is not a string")
if "syringe" in formatted.lower():
formatted = "⚠️ SAFETY ALERT: Rejected dangerous suggestion\n\n" + formatted
return formatted
except Exception as e:
logger.error(f"Groq formatting failed: {str(e)}")
raise ValueError(f"Groq processing failed: {str(e)}")
def _generate_with_groq(self, query: str, context: str = None, refined_prompt: str = None) -> str:
"""
Generate medical response using Groq with two modes:
1. Direct mode (query + context)
2. Refined prompt mode (FLAN-processed prompt)
Args:
query: Original user query
context: Optional RAG context
refined_prompt: Optional FLAN-processed prompt
"""
def _get_groq_client():
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
# Verify Groq client
if not self.groq_client:
self.groq_client = _get_groq_client()
if not self.groq_client:
raise ValueError("Groq client not available")
try:
# Determine which mode to use
if refined_prompt:
# Refined prompt mode (RAG→FLAN→Groq pipeline)
messages = [
{
"role": "system",
"content": (
f"""{self.system_prompt}\n
Your task is to expand the medical information into comprehensive,
descriptive guidance while preserving all safety considerations.
You are a WHO medical advisor for Gaza. Provide 1) Extremely detailed step-by-step guidance
2) Multiple treatment options for different resource scenarios
3)Clear timeframes for each action
4)Islamic medical considerations
5)Format with emoji section headers
"""
)
},
{
"role": "user",
"content": f"""Expand this into comprehensive medical guidance:
{refined_prompt}
Include:
1. Detailed procedural steps
2. Pain management techniques
3. Infection prevention measures
4. When to seek emergency care"""
}
]
max_tokens = 2000 # Allow longer responses for descriptive answers
temperature = 0.5 # Slightly higher for creativity
else:
# Direct mode (query + context)
messages = [
{
"role": "system",
"content": (
f"{self.system_prompt}\n"
"Provide detailed 5-7 step guidance when applicable."
)
},
{
"role": "user",
"content": (
f"Query: {query}\n"
f"Context: {context if context else 'No additional context'}\n\n"
"Provide comprehensive guidance with:\n"
"1. Detailed steps\n2. Alternative methods\n3. Gaza-specific adaptations"
)
}
]
max_tokens = 1500 # Slightly shorter for direct responses
temperature = 0.4 # Balanced between creativity and accuracy
# Make the API call
response = self.groq_client.chat.completions.create(
model="llama3-70b-8192", # Using latest model
messages=messages,
temperature=temperature,
max_tokens=max_tokens,
top_p=0.9
)
result = response.choices[0].message.content
if not result:
raise ValueError("Empty response from Groq")
# Post-processing steps
if refined_prompt and "Context References:" in refined_prompt:
# Preserve RAG references in refined prompt mode
refs = refined_prompt.split("Context References:")[1]
result = f"{result}\n\n📚 Source References:{refs}"
elif context:
# Add basic context reference in direct mode
result += "\n\nℹ️ Context: Based on verified medical sources"
# Safety checks (applied to both modes)
if "syringe" in result.lower():
result = "⚠️ SAFETY ALERT: Rejected dangerous suggestion\n\n" + result
if "Gaza Red Crescent" not in result:
result += "\n\n📞 Verify with Gaza Red Crescent (101) if condition worsens"
return result
except Exception as e:
logger.error(f"Groq generation failed: {str(e)}")
raise ValueError(f"Groq processing error: {str(e)}")
def _generate_response(self, query: str, context: str) -> str:
"""Enhanced RAG → FLAN → Groq pipeline with fallbacks"""
# FIXED: Removed early return that was bypassing FLAN→Groq pipeline
def _get_groq_client():
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
# 1. Get broader RAG context (3 results instead of 1)
rag_results = self.knowledge_base.search(query, k=6)
context = self._prepare_context(rag_results) # ✅ FIXED: ensure context exists
top_score = rag_results[0]['score'] if rag_results else 0
logger.info(f"Found {len(rag_results)} RAG results (top score: {rag_results[0]['score'] if rag_results else 0:.2f})")
# 2. Try RAG→FLAN→Groq pipeline
try:
# Create refined prompt using FLAN
refined_prompt = self._create_prompt_from_rag(query, rag_results)
logger.info(f"Refined prompt: {refined_prompt[:100]}...")
# Generate with Groq if available
if hasattr(self, 'groq_client') and self.groq_client:
try:
groq_response = self._generate_with_groq(query=query, refined_prompt=refined_prompt)
if groq_response:
return groq_response
except Exception as groq_error:
logger.warning(f"Groq generation failed: {str(groq_error)}")
except Exception as pipe_error:
logger.warning(f"RAG→FLAN→Groq pipeline failed: {str(pipe_error)}")
# 3. Fallback to direct FLAN generation
if self.llm and self.tokenizer:
try:
# Use the original context (not refined prompt) for fallback
flan_response = self._generate_with_flan(query, context)
if flan_response:
return flan_response
except Exception as flan_error:
logger.error(f"FLAN generation failed: {str(flan_error)}")
# 4. Ultimate fallback to cached knowledge
if rag_results:
return self._format_kb_response(rag_results[0]['text'])
# 5. Final emergency fallback
return self._generate_fallback_response(query, context)
def _format_final_response(self, response: str) -> str:
"""Ensure response meets all Gaza-specific requirements"""
clean_response = response.split("CONTEXT:")[0].strip()
for icon in ["🩹", "💡", "⚠️"]:
if icon not in clean_response:
clean_response = clean_response.replace("Immediate Actions", f"Immediate Actions {icon}", 1)
break
if "📍 Gaza Context:" not in clean_response:
clean_response += "\n\n📍 Gaza Context: This guidance considers resource limitations. Adapt based on available supplies and seek professional medical care when accessible."
return clean_response
def _get_error_response(self, query: str, error: Exception) -> str:
"""User-friendly error message with Gaza contacts"""
return f"""⚠️ We're unable to process your query about "{query}"
For immediate medical assistance:
📞 Palestinian Red Crescent: 101
📞 Civil Defense: 102
(Technical issue: {str(error)}...)"""
def _generate_fallback_response(self, query: str, context: str) -> str:
"""Enhanced fallback response with Gaza-specific guidance"""
gaza_guidance = {
"burn": "For burns: Use clean, cool water if available. If water is scarce, use clean cloth. Avoid ice. Seek medical help urgently.",
"bleeding": "For bleeding: Apply direct pressure with clean cloth. Elevate if possible. If severe, seek immediate medical attention.",
"wound": "For wounds: Clean hands if possible. Apply pressure to stop bleeding. Cover with clean material. Watch for infection signs.",
"infection": "Signs of infection: Redness, warmth, swelling, pus, fever. Seek medical care immediately if available.",
"pain": "For pain management: Rest, elevation, cold/warm compress as appropriate. Avoid aspirin in children."
}
query_lower = query.lower()
for condition, guidance in gaza_guidance.items():
if condition in query_lower:
return f"{guidance}\n\nContext from medical sources:\n{context}..."
return f"Medical guidance for: {query}\n\nGeneral advice: Prioritize safety, seek professional help when available, consider resource limitations in Gaza.\n\nRelevant information:\n{context[:600]}..."
def _prepare_final_response(
self,
response: str,
search_results: List[Dict[str, Any]],
safety_check: Dict[str, Any],
response_time: float
) -> Dict[str, Any]:
def _get_groq_client():
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
# Ensure response is a string
if response is None:
response = "Unable to generate response. Please try again."
elif not isinstance(response, str):
response = str(response)
"""Enhanced final response preparation with more metadata"""
# Ensure response is a string
if not isinstance(response, str):
response = "Unable to generate response. Please try again."
# Ensure safety_check has required fields
if not isinstance(safety_check, dict):
safety_check = {
"confidence_score": 0.5,
"issues": [],
"warnings": ["Response validation failed"],
"is_safe": False
}
# Add safety warnings if needed
if not safety_check["is_safe"]:
response = f"⚠️ MEDICAL CAUTION: {response}\n\n🚨 Please verify this guidance with a medical professional when possible."
if safety_check["is_safe"] and hasattr(self, 'groq_client') and self.groq_client:
try:
enhanced = self._enhance_response_with_groq(response, search_results)
if enhanced:
response = enhanced
except Exception as e:
logger.warning(f"Response enhancement failed: {e}")
# Add Gaza-specific disclaimer
# Extract unique sources
sources = list(set(res.get("source", "unknown") for res in search_results)) if search_results else []
# Calculate confidence based on multiple factors
base_confidence = safety_check.get("confidence_score", 0.5)
context_bonus = 0.1 if search_results else 0.0
safety_penalty = 0.2 if not safety_check.get("is_safe", True) else 0.0
final_confidence = max(0.0, min(1.0, base_confidence + context_bonus - safety_penalty))
return {
"response": response,
"confidence": final_confidence,
"sources": sources,
"search_results_count": len(search_results),
"safety_issues": safety_check.get("issues", []),
"safety_warnings": safety_check.get("warnings", []),
"response_time": round(response_time, 2),
"timestamp": datetime.now().isoformat()[:19],
"cached": False
}
if not hasattr(self, 'groq_client') or not self.groq_client:
return response
try:
groq_client = self._get_groq_client()
models = groq_client.models.list()
logger.info("🧪 Available Groq models:")
for m in models.data:
logger.info(f" - {m.id}")
messages = [
{"role": "system", "content": self.system_prompt},
{"role": "user", "content": f"Enhance this medical response:\n\n{response}"}
]
enhanced = self.groq_client.chat.completions.create(
model="deepseek-r1-distill-llama-70b",
messages=messages,
temperature=0.3,
max_tokens=1000
)
if enhanced and enhanced.choices:
return enhanced.choices[0].message.content
return response
except Exception as e:
logger.warning(f"Response enhancement failed: {e}")
return response
def _enhance_response_with_groq(self, response: str, search_results: List[Dict[str, Any]]) -> str:
"""Enhance response using Groq's capabilities"""
def _get_groq_client():
from groq import Groq
return Groq(api_key=os.getenv("GROQ_API_KEY"))
if not hasattr(self, 'groq_client') or not self.groq_client:
return response
try:
messages = [
{"role": "system", "content": self.system_prompt},
{"role": "user", "content": f"Enhance this medical response:\n\n{response}"}
]
enhanced = self.groq_client.chat.completions.create(
model="llama3-70b-8192", # Updated model name
messages=messages,
temperature=0.3,
max_tokens=2000
)
if enhanced and enhanced.choices:
return enhanced.choices[0].message.content
return response
except Exception as e:
logger.warning(f"Response enhancement failed: {e}")
return response
def _create_error_response(self, error_msg: str) -> Dict[str, Any]:
"""Enhanced error response with helpful information"""
return {
"response": f"⚠️ System Error: Unable to process your medical query at this time.\n\nError: {error_msg}\n\n🚨 For immediate medical emergencies, seek professional help directly.\n\n📞 Gaza Emergency Numbers:\n- Palestinian Red Crescent: 101\n- Civil Defense: 102",
"confidence": 0.0,
"sources": [],
"search_results_count": 0,
"safety_issues": ["System error occurred"],
"safety_warnings": ["Unable to validate medical accuracy "],
"response_time": 0.0,
"timestamp": datetime.now().isoformat()[:19],
"cached": False,
"error": True
}
|