Blaiseboy commited on
Commit
ea9b146
·
verified ·
1 Parent(s): 7242cdd

Delete medical_chatbot.py

Browse files
Files changed (1) hide show
  1. medical_chatbot.py +0 -608
medical_chatbot.py DELETED
@@ -1,608 +0,0 @@
1
- # Setup and Installation
2
-
3
- import torch
4
- print("🖥️ System Check:")
5
- print(f"CUDA available: {torch.cuda.is_available()}")
6
- if torch.cuda.is_available():
7
- print(f"GPU device: {torch.cuda.get_device_name(0)}")
8
- print(f"GPU memory: {torch.cuda.get_device_properties(0).total_memory / 1e9:.1f} GB")
9
- else:
10
- print("⚠️ No GPU detected - BioGPT will run on CPU")
11
-
12
- print("\n🔧 Loading required packages...")
13
-
14
- # Import Libraries
15
-
16
- import os
17
- import re
18
- import torch
19
- import warnings
20
- import numpy as np
21
- import faiss # FAISS for vector search
22
- from transformers import (
23
- AutoTokenizer,
24
- AutoModelForCausalLM,
25
- pipeline,
26
- BitsAndBytesConfig
27
- )
28
- from sentence_transformers import SentenceTransformer
29
- from typing import List, Dict, Optional
30
- import time
31
- from datetime import datetime
32
- import json
33
- import pickle
34
-
35
- # Suppress warnings for cleaner output
36
- warnings.filterwarnings('ignore')
37
-
38
- print("📚 Libraries imported successfully!")
39
- print(f"🔍 FAISS version: {faiss.__version__}")
40
- print("🎯 Using FAISS for vector search")
41
-
42
- # BioGPT Medical Chatbot Class
43
-
44
- class ColabBioGPTChatbot:
45
- def __init__(self, use_gpu=True, use_8bit=True):
46
- """Initialize BioGPT chatbot optimized for deployment"""
47
- print("🏥 Initializing Professional BioGPT Medical Chatbot...")
48
-
49
- # Force CPU for HF Spaces if needed
50
- self.device = "cuda" if torch.cuda.is_available() and use_gpu else "cpu"
51
- self.use_8bit = use_8bit and torch.cuda.is_available()
52
-
53
- print(f"🖥️ Using device: {self.device}")
54
- if self.use_8bit:
55
- print("💾 Using 8-bit quantization for memory efficiency")
56
-
57
- # Setup components
58
- self.setup_embeddings()
59
- self.setup_faiss_index()
60
- self.setup_biogpt()
61
-
62
- # Conversation tracking
63
- self.conversation_history = []
64
- self.knowledge_chunks = []
65
-
66
- print("✅ BioGPT Medical Chatbot ready for professional medical assistance!")
67
-
68
- def setup_embeddings(self):
69
- """Setup medical-optimized embeddings"""
70
- print("🔧 Loading medical embeddings...")
71
- try:
72
- # Use a smaller, more efficient model for deployment
73
- self.embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
74
- self.embedding_dim = self.embedding_model.get_sentence_embedding_dimension()
75
- print(f"✅ Embeddings loaded (dimension: {self.embedding_dim})")
76
- self.use_embeddings = True
77
- except Exception as e:
78
- print(f"⚠️ Embeddings failed: {e}")
79
- self.embedding_model = None
80
- self.embedding_dim = 384
81
- self.use_embeddings = False
82
-
83
- def setup_faiss_index(self):
84
- """Setup faiss for CPU-based vector search"""
85
- print("🔧 Setting up FAISS vector database...")
86
- try:
87
- print('Using CPU FAISS index for maximum compatibility')
88
- self.faiss_index = faiss.IndexFlatIP(self.embedding_dim)
89
- self.use_gpu_faiss = False
90
- self.faiss_ready = True
91
- self.collection = self.faiss_index
92
- print("✅ FAISS CPU index initialized successfully")
93
- except Exception as e:
94
- print(f"❌ FAISS setup failed: {e}")
95
- self.faiss_index = None
96
- self.faiss_ready = False
97
- self.collection = None
98
-
99
- def setup_biogpt(self):
100
- """Setup BioGPT model with optimizations for deployment"""
101
- print("🧠 Loading BioGPT model...")
102
-
103
- # Try BioGPT first, fallback to smaller models if needed
104
- model_options = [
105
- "microsoft/BioGPT-Large",
106
- "microsoft/BioGPT", # Smaller version
107
- "microsoft/DialoGPT-medium", # Fallback
108
- "gpt2" # Final fallback
109
- ]
110
-
111
- for model_name in model_options:
112
- try:
113
- print(f" Attempting to load: {model_name}")
114
-
115
- # Setup quantization config for memory efficiency
116
- if self.use_8bit and "BioGPT" in model_name:
117
- quantization_config = BitsAndBytesConfig(
118
- load_in_8bit=True,
119
- llm_int8_threshold=6.0,
120
- llm_int8_has_fp16_weight=False,
121
- )
122
- else:
123
- quantization_config = None
124
-
125
- # Load tokenizer
126
- self.tokenizer = AutoTokenizer.from_pretrained(model_name)
127
-
128
- # Set padding token
129
- if self.tokenizer.pad_token is None:
130
- self.tokenizer.pad_token = self.tokenizer.eos_token
131
-
132
- # Load model with proper settings for deployment
133
- start_time = time.time()
134
-
135
- model_kwargs = {
136
- "torch_dtype": torch.float16 if self.device == "cuda" else torch.float32,
137
- "trust_remote_code": True,
138
- "low_cpu_mem_usage": True, # Important for deployment
139
- }
140
-
141
- if quantization_config:
142
- model_kwargs["quantization_config"] = quantization_config
143
- model_kwargs["device_map"] = "auto"
144
-
145
- self.model = AutoModelForCausalLM.from_pretrained(
146
- model_name,
147
- **model_kwargs
148
- )
149
-
150
- # Move to device if not using device_map
151
- if self.device == "cuda" and quantization_config is None:
152
- self.model = self.model.to(self.device)
153
-
154
- load_time = time.time() - start_time
155
- print(f"✅ {model_name} loaded successfully! ({load_time:.1f} seconds)")
156
-
157
- # Test the model
158
- self.test_model()
159
- break # Success, exit the loop
160
-
161
- except Exception as e:
162
- print(f"❌ {model_name} loading failed: {e}")
163
- if model_name == model_options[-1]: # Last option failed
164
- print("❌ All models failed to load")
165
- self.model = None
166
- self.tokenizer = None
167
- continue
168
-
169
- def test_model(self):
170
- """Test the loaded model with a simple query"""
171
- print("🧪 Testing model...")
172
- try:
173
- test_prompt = "Fever in children can be caused by"
174
- inputs = self.tokenizer(test_prompt, return_tensors="pt")
175
-
176
- if self.device == "cuda":
177
- inputs = {k: v.to(self.device) for k, v in inputs.items()}
178
-
179
- with torch.no_grad():
180
- outputs = self.model.generate(
181
- **inputs,
182
- max_new_tokens=20,
183
- do_sample=True,
184
- temperature=0.7,
185
- pad_token_id=self.tokenizer.eos_token_id
186
- )
187
-
188
- response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
189
- print(f"✅ Model test successful!")
190
- print(f" Test response: {response}")
191
-
192
- except Exception as e:
193
- print(f"⚠️ Model test failed: {e}")
194
-
195
- def load_medical_data(self, file_path: str):
196
- """Load and process medical data with progress tracking"""
197
- print(f"📖 Loading medical data from {file_path}...")
198
-
199
- try:
200
- with open(file_path, 'r', encoding='utf-8') as f:
201
- text = f.read()
202
- print(f"📄 File loaded: {len(text):,} characters")
203
- except FileNotFoundError:
204
- print(f"❌ File {file_path} not found!")
205
- return False
206
- except Exception as e:
207
- print(f"❌ Error loading file: {e}")
208
- return False
209
-
210
- # Create chunks optimized for medical content
211
- print("📝 Creating medical-optimized chunks...")
212
- chunks = self.create_medical_chunks(text)
213
- print(f"📋 Created {len(chunks)} medical chunks")
214
-
215
- self.knowledge_chunks = chunks
216
-
217
- # Generate embeddings with progress and add to FAISS index
218
- if self.use_embeddings and self.embedding_model and self.faiss_ready:
219
- return self.generate_embeddings_with_progress(chunks)
220
-
221
- print("✅ Medical data loaded (text search mode)")
222
- return True
223
-
224
- def create_medical_chunks(self, text: str, chunk_size: int = 400) -> List[Dict]:
225
- """Create medically-optimized text chunks"""
226
- chunks = []
227
-
228
- # Split by medical sections first
229
- medical_sections = self.split_by_medical_sections(text)
230
-
231
- chunk_id = 0
232
- for section in medical_sections:
233
- if len(section.split()) > chunk_size:
234
- # Split large sections by sentences
235
- sentences = re.split(r'[.!?]+', section)
236
- current_chunk = ""
237
-
238
- for sentence in sentences:
239
- sentence = sentence.strip()
240
- if not sentence:
241
- continue
242
-
243
- if len(current_chunk.split()) + len(sentence.split()) < chunk_size:
244
- current_chunk += sentence + ". "
245
- else:
246
- if current_chunk.strip():
247
- chunks.append({
248
- 'id': chunk_id,
249
- 'text': current_chunk.strip(),
250
- 'medical_focus': self.identify_medical_focus(current_chunk)
251
- })
252
- chunk_id += 1
253
- current_chunk = sentence + ". "
254
-
255
- if current_chunk.strip():
256
- chunks.append({
257
- 'id': chunk_id,
258
- 'text': current_chunk.strip(),
259
- 'medical_focus': self.identify_medical_focus(current_chunk)
260
- })
261
- chunk_id += 1
262
- else:
263
- chunks.append({
264
- 'id': chunk_id,
265
- 'text': section,
266
- 'medical_focus': self.identify_medical_focus(section)
267
- })
268
- chunk_id += 1
269
-
270
- return chunks
271
-
272
- def split_by_medical_sections(self, text: str) -> List[str]:
273
- """Split text by medical sections"""
274
- # Look for medical section headers
275
- section_patterns = [
276
- r'\n\s*(?:SYMPTOMS?|TREATMENT|DIAGNOSIS|CAUSES?|PREVENTION|MANAGEMENT).*?\n',
277
- r'\n\s*\d+\.\s+', # Numbered sections
278
- r'\n\n+' # Paragraph breaks
279
- ]
280
-
281
- sections = [text]
282
- for pattern in section_patterns:
283
- new_sections = []
284
- for section in sections:
285
- splits = re.split(pattern, section, flags=re.IGNORECASE)
286
- new_sections.extend([s.strip() for s in splits if len(s.strip()) > 100])
287
- sections = new_sections
288
-
289
- return sections
290
-
291
- def identify_medical_focus(self, text: str) -> str:
292
- """Identify the medical focus of a text chunk"""
293
- text_lower = text.lower()
294
-
295
- # Medical categories
296
- categories = {
297
- 'pediatric_symptoms': ['fever', 'cough', 'rash', 'vomiting', 'diarrhea'],
298
- 'treatments': ['treatment', 'therapy', 'medication', 'antibiotics'],
299
- 'diagnosis': ['diagnosis', 'diagnostic', 'symptoms', 'signs'],
300
- 'emergency': ['emergency', 'urgent', 'serious', 'hospital'],
301
- 'prevention': ['prevention', 'vaccine', 'immunization', 'avoid']
302
- }
303
-
304
- for category, keywords in categories.items():
305
- if any(keyword in text_lower for keyword in keywords):
306
- return category
307
-
308
- return 'general_medical'
309
-
310
- def generate_embeddings_with_progress(self, chunks: List[Dict]) -> bool:
311
- """Generate embeddings with progress tracking and add to FAISS index"""
312
- print("🔮 Generating medical embeddings and adding to FAISS index...")
313
-
314
- if not self.embedding_model or not self.faiss_index:
315
- print("❌ Embedding model or FAISS index not available.")
316
- return False
317
-
318
- try:
319
- texts = [chunk['text'] for chunk in chunks]
320
-
321
- # Generate embeddings in batches with progress
322
- batch_size = 32
323
- all_embeddings = []
324
-
325
- for i in range(0, len(texts), batch_size):
326
- batch_texts = texts[i:i+batch_size]
327
- batch_embeddings = self.embedding_model.encode(batch_texts, show_progress_bar=False)
328
- all_embeddings.extend(batch_embeddings)
329
-
330
- # Show progress
331
- progress = min(i + batch_size, len(texts))
332
- print(f" Progress: {progress}/{len(texts)} chunks processed", end='\r')
333
-
334
- print(f"\n ✅ Generated embeddings for {len(texts)} chunks")
335
-
336
- # Add embeddings to FAISS index
337
- print("💾 Adding embeddings to FAISS index...")
338
- self.faiss_index.add(np.array(all_embeddings))
339
-
340
- print("✅ Medical embeddings added to FAISS index successfully!")
341
- return True
342
-
343
- except Exception as e:
344
- print(f"❌ Embedding generation or FAISS add failed: {e}")
345
- return False
346
-
347
- def retrieve_medical_context(self, query: str, n_results: int = 3) -> List[str]:
348
- """Retrieve relevant medical context using embeddings or keyword search"""
349
- if self.use_embeddings and self.embedding_model and self.faiss_ready:
350
- try:
351
- # Generate query embedding
352
- query_embedding = self.embedding_model.encode([query])
353
-
354
- # Search for similar content in FAISS index
355
- distances, indices = self.faiss_index.search(np.array(query_embedding), n_results)
356
-
357
- # Retrieve the corresponding chunks
358
- context_chunks = [self.knowledge_chunks[i]['text'] for i in indices[0] if i != -1]
359
-
360
- if context_chunks:
361
- return context_chunks
362
-
363
- except Exception as e:
364
- print(f"⚠️ Embedding search failed: {e}")
365
-
366
- # Fallback to keyword search
367
- print("⚠️ Falling back to keyword search.")
368
- return self.keyword_search_medical(query, n_results)
369
-
370
- def keyword_search_medical(self, query: str, n_results: int) -> List[str]:
371
- """Medical-focused keyword search"""
372
- if not self.knowledge_chunks:
373
- return []
374
-
375
- query_words = set(query.lower().split())
376
- chunk_scores = []
377
-
378
- for chunk_info in self.knowledge_chunks:
379
- chunk_text = chunk_info['text']
380
- chunk_words = set(chunk_text.lower().split())
381
-
382
- # Calculate relevance score
383
- word_overlap = len(query_words.intersection(chunk_words))
384
- base_score = word_overlap / len(query_words) if query_words else 0
385
-
386
- # Boost medical content
387
- medical_boost = 0
388
- if chunk_info.get('medical_focus') in ['pediatric_symptoms', 'treatments', 'diagnosis']:
389
- medical_boost = 0.5
390
-
391
- final_score = base_score + medical_boost
392
-
393
- if final_score > 0:
394
- chunk_scores.append((final_score, chunk_text))
395
-
396
- # Return top matches
397
- chunk_scores.sort(reverse=True)
398
- return [chunk for _, chunk in chunk_scores[:n_results]]
399
-
400
- def generate_biogpt_response(self, context: str, query: str) -> str:
401
- """Generate medical response using BioGPT only"""
402
- if not self.model or not self.tokenizer:
403
- return "⚠️ Medical AI model not available. This chatbot requires BioGPT for accurate medical information. Please check the setup or try restarting."
404
-
405
- try:
406
- # Create medical-focused prompt
407
- prompt = f"""Medical Context: {context[:800]}
408
-
409
- Question: {query}
410
-
411
- Medical Answer:"""
412
-
413
- # Tokenize input
414
- inputs = self.tokenizer(
415
- prompt,
416
- return_tensors="pt",
417
- truncation=True,
418
- max_length=1024
419
- )
420
-
421
- # Move inputs to the correct device
422
- if self.device == "cuda":
423
- inputs = {k: v.to(self.device) for k, v in inputs.items()}
424
-
425
- # Generate response
426
- with torch.no_grad():
427
- outputs = self.model.generate(
428
- **inputs,
429
- max_new_tokens=150,
430
- do_sample=True,
431
- temperature=0.7,
432
- top_p=0.9,
433
- pad_token_id=self.tokenizer.eos_token_id,
434
- repetition_penalty=1.1
435
- )
436
-
437
- # Decode response
438
- full_response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
439
-
440
- # Extract just the generated part
441
- if "Medical Answer:" in full_response:
442
- generated_response = full_response.split("Medical Answer:")[-1].strip()
443
- else:
444
- generated_response = full_response[len(prompt):].strip()
445
-
446
- # Clean up response
447
- cleaned_response = self.clean_medical_response(generated_response)
448
-
449
- return cleaned_response
450
-
451
- except Exception as e:
452
- print(f"⚠️ BioGPT generation failed: {e}")
453
- return "⚠️ Unable to generate medical response. The medical AI model encountered an error. Please try rephrasing your question or contact support."
454
-
455
- def clean_medical_response(self, response: str) -> str:
456
- """Clean and format medical response"""
457
- # Remove incomplete sentences and limit length
458
- sentences = re.split(r'[.!?]+', response)
459
- clean_sentences = []
460
-
461
- for sentence in sentences:
462
- sentence = sentence.strip()
463
- if len(sentence) > 10 and not sentence.endswith(('and', 'or', 'but', 'however')):
464
- clean_sentences.append(sentence)
465
- if len(clean_sentences) >= 3: # Limit to 3 sentences
466
- break
467
-
468
- if clean_sentences:
469
- cleaned = '. '.join(clean_sentences) + '.'
470
- else:
471
- cleaned = response[:200] + '...' if len(response) > 200 else response
472
-
473
- return cleaned
474
-
475
- def fallback_response(self, context: str, query: str) -> str:
476
- """Fallback response when BioGPT fails"""
477
- # Extract key sentences from context
478
- sentences = [s.strip() for s in context.split('.') if len(s.strip()) > 20]
479
-
480
- if sentences:
481
- response = sentences[0] + '.'
482
- if len(sentences) > 1:
483
- response += ' ' + sentences[1] + '.'
484
- else:
485
- response = context[:300] + '...'
486
-
487
- return response
488
-
489
- def handle_conversational_interactions(self, query: str) -> Optional[str]:
490
- """Handle comprehensive conversational interactions"""
491
- query_lower = query.lower().strip()
492
-
493
- # Use more specific patterns for greetings
494
- greeting_patterns = [
495
- r'^\s*(hello|hi|hey|hiya|howdy)\s*$',
496
- r'^\s*(good morning|good afternoon|good evening|good day)\s*$',
497
- r'^\s*(what\'s up|whats up|sup|yo)\s*$',
498
- r'^\s*(greetings|salutations)\s*$',
499
- r'^\s*(how are you|how are you doing|how\'s it going|hows it going)\s*$',
500
- r'^\s*(good to meet you|nice to meet you|pleased to meet you)\s*$'
501
- ]
502
-
503
- for pattern in greeting_patterns:
504
- if re.match(pattern, query_lower):
505
- responses = [
506
- "👋 Hello! I'm BioGPT, your professional medical AI assistant specialized in pediatric medicine. I'm here to provide evidence-based medical information. What health concern can I help you with today?",
507
- "🏥 Hi there! I'm a medical AI assistant powered by BioGPT, trained on medical literature. I can help answer questions about children's health and medical conditions. How can I assist you?",
508
- "👋 Greetings! I'm your AI medical consultant, ready to help with pediatric health questions using the latest medical knowledge. What would you like to know about?"
509
- ]
510
- return np.random.choice(responses)
511
-
512
- # Handle thanks and other conversational patterns...
513
- # (keeping the rest of the conversational handling as before)
514
-
515
- # Return None if no conversational pattern matches
516
- return None
517
-
518
- def chat(self, query: str) -> str:
519
- """Main chat function with BioGPT medical-only responses"""
520
- if not query.strip():
521
- return "Hello! I'm BioGPT, your professional medical AI assistant. How can I help you with pediatric medical questions today?"
522
-
523
- # Handle comprehensive conversational interactions first
524
- conversational_response = self.handle_conversational_interactions(query)
525
- if conversational_response:
526
- # Add to conversation history
527
- self.conversation_history.append({
528
- 'query': query,
529
- 'response': conversational_response,
530
- 'timestamp': datetime.now().isoformat(),
531
- 'type': 'conversational'
532
- })
533
- return conversational_response
534
-
535
- # Check if medical model is available
536
- if not self.model or not self.tokenizer:
537
- return "⚠️ **Medical AI Unavailable**: This chatbot requires BioGPT for accurate medical information. The medical model failed to load. Please contact support or try restarting the application."
538
-
539
- if not self.knowledge_chunks:
540
- return "Please load medical data first to access the medical knowledge base."
541
-
542
- print(f"🔍 Processing medical query: {query}")
543
-
544
- # Retrieve relevant medical context using FAISS or keyword search
545
- start_time = time.time()
546
- context = self.retrieve_medical_context(query)
547
- retrieval_time = time.time() - start_time
548
-
549
- if not context:
550
- return "I don't have specific information about this topic in my medical database. Please consult with a healthcare professional for personalized medical advice."
551
-
552
- print(f" 📚 Context retrieved ({retrieval_time:.2f}s)")
553
-
554
- # Generate response with BioGPT
555
- start_time = time.time()
556
- main_context = '\n\n'.join(context)
557
- response = self.generate_biogpt_response(main_context, query)
558
- generation_time = time.time() - start_time
559
-
560
- print(f" 🧠 Response generated ({generation_time:.2f}s)")
561
-
562
- # Format final response
563
- final_response = f"🩺 **Medical Information:** {response}\n\n⚠️ **Important:** This information is for educational purposes only. Always consult with qualified healthcare professionals for medical diagnosis, treatment, and personalized advice."
564
-
565
- # Add to conversation history
566
- self.conversation_history.append({
567
- 'query': query,
568
- 'response': final_response,
569
- 'timestamp': datetime.now().isoformat(),
570
- 'retrieval_time': retrieval_time,
571
- 'generation_time': generation_time,
572
- 'type': 'medical'
573
- })
574
-
575
- return final_response
576
-
577
- def get_conversation_summary(self) -> Dict:
578
- """Get conversation statistics"""
579
- if not self.conversation_history:
580
- return {"message": "No conversations yet"}
581
-
582
- # Filter medical conversations for performance stats
583
- medical_conversations = [h for h in self.conversation_history if h.get('type') == 'medical']
584
-
585
- if not medical_conversations:
586
- return {
587
- "total_conversations": len(self.conversation_history),
588
- "medical_conversations": 0,
589
- "conversational_interactions": len(self.conversation_history),
590
- "model_info": "BioGPT" if self.model and "BioGPT" in str(type(self.model)) else "Fallback Model",
591
- "vector_search": "FAISS CPU" if self.faiss_ready else "Keyword Search",
592
- "device": self.device
593
- }
594
-
595
- avg_retrieval_time = sum(h.get('retrieval_time', 0) for h in medical_conversations) / len(medical_conversations)
596
- avg_generation_time = sum(h.get('generation_time', 0) for h in medical_conversations) / len(medical_conversations)
597
-
598
- return {
599
- "total_conversations": len(self.conversation_history),
600
- "medical_conversations": len(medical_conversations),
601
- "conversational_interactions": len(self.conversation_history) - len(medical_conversations),
602
- "avg_retrieval_time": f"{avg_retrieval_time:.2f}s",
603
- "avg_generation_time": f"{avg_generation_time:.2f}s",
604
- "model_info": "BioGPT" if self.model and "BioGPT" in str(type(self.model)) else "Fallback Model",
605
- "vector_search": "FAISS CPU" if self.faiss_ready else "Keyword Search",
606
- "device": self.device,
607
- "quantization": "8-bit" if self.use_8bit else "16-bit/32-bit"
608
- }