added new model

app.py

@@ -100,6 +100,10 @@ def generate_response(prompt, max_tokens=None, temperature=None, top_p=None):
     """Generate response using the loaded model"""
     global model, tokenizer, current_model_name
     
+    print(f"Starting generation for prompt: {prompt}")
+    if not prompt or not prompt.strip():
+        return "Please enter a question. 😊"
+    
     if model is None or tokenizer is None:
         return "❌ Model not loaded. Please wait for initialization or try restarting the space."
     
@@ -108,14 +112,9 @@ def generate_response(prompt, max_tokens=None, temperature=None, top_p=None):
     temperature = temperature or GENERATION_DEFAULTS["temperature"]
     top_p = top_p or GENERATION_DEFAULTS["top_p"]
     
-    print(f"🔄 Starting generation for prompt: {prompt[:50]}{'...' if len(prompt) > 50 else ''}")
-    if not prompt or not prompt.strip():
-        return "Please enter a question. 😊"
-    print(f"📊 Generation params: max_tokens={max_tokens}, temp={temperature}, top_p={top_p}")
-    
     try:
-        full_prompt = f"{MEDICAL_SYSTEM_PROMPT}\n\nPatient/User: {prompt}\nMedical Assistant:"
-        print(f"📝 Using full prompt: {full_prompt[:100]}{'...' if len(full_prompt) > 100 else ''}")
+        full_prompt = f"{MEDICAL_SYSTEM_PROMPT}\n\nPatient/User: {prompt}\n"
+        print(f"Full prompt: {full_prompt}")
         
         # Tokenize input with proper truncation
         inputs = tokenizer(
@@ -130,20 +129,18 @@ def generate_response(prompt, max_tokens=None, temperature=None, top_p=None):
         device = next(model.parameters()).device
         inputs = {k: v.to(device) for k, v in inputs.items()}
         
-        # Generation parameters (optimized for CPU performance)
+        # Generation parameters
         generation_kwargs = {
-            "max_new_tokens": min(max_tokens, 100),  # Further reduced for better control
+            "max_new_tokens": min(max_tokens, 1024),
             "temperature": temperature,
             "top_p": top_p,
             "do_sample": GENERATION_DEFAULTS["do_sample"],
             "pad_token_id": tokenizer.eos_token_id,
             "repetition_penalty": GENERATION_DEFAULTS["repetition_penalty"],
-            "no_repeat_ngram_size": GENERATION_DEFAULTS["no_repeat_ngram_size"],
-            "early_stopping": True,  # Stop early when end token is found
-            "num_beams": 1  # Use greedy decoding for speed on CPU
+            "no_repeat_ngram_size": GENERATION_DEFAULTS["no_repeat_ngram_size"]
         }
         
-        print(f"🔧 Generating with kwargs: {generation_kwargs}")
+        print(f"Generating with kwargs: {generation_kwargs}")
         
         # Generate response
         print(f"🤖 Generating response with {current_model_name}...")
@@ -156,19 +153,14 @@ def generate_response(prompt, max_tokens=None, temperature=None, top_p=None):
         generation_time = time.time() - start_time
         print(f"⏱️ Generation completed in {generation_time:.2f} seconds")
         
-        # Decode response
+        # Decode response and extract new content
         full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        response = full_response.replace(full_prompt, "").strip()
+        print(f"Generated response: {response}")
         
-        # Extract only the new generated text
-        if "Medical Assistant:" in full_response:
-            response = full_response.split("Medical Assistant:")[-1].strip()
-        else:
-            # Fallback extraction
-            response = full_response[len(full_prompt):].strip()
-        
-        # Clean up response - keep it natural as per CareConnect guidelines
-        if not response or len(response.strip()) < 3:
-            response = "Could you please provide more details about your question so I can help you better? 😊"
+        # Clean up response
+        if not response or len(response.strip()) < 10:
+            response = "Sorry, I couldn't process that. Try again or see a doctor. 😊"
         
         print(f"✅ Generated response length: {len(response)} characters")
         print(f"📄 Response preview: {response[:150]}{'...' if len(response) > 150 else ''}")

app_flan_t5.py

@@ -0,0 +1,228 @@
+import gradio as gr
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+import logging
+import gc
+import warnings
+import os
+from huggingface_hub import login
+
+# Login with the secret token
+login(token=os.getenv("HF_TOKEN"))
+
+# Suppress warnings
+warnings.filterwarnings("ignore")
+logging.getLogger("transformers").setLevel(logging.ERROR)
+
+# Configuration for FLAN-T5 Small
+MODEL_NAME = "google/flan-t5-small"
+MAX_NEW_TOKENS = 100
+TEMPERATURE = 0.7
+TOP_P = 0.9
+
+# Medical system prompt optimized for FLAN-T5
+MEDICAL_SYSTEM_PROMPT = """You are a friendly and smart medical assistant. Your job is to give short, clear, and helpful health information.
+
+Your answers should:
+- Stay focused. No long essays or extra fluff.
+- Give basic helpful steps for common symptoms like fever, cough, or headache (e.g., rest, drink fluids, take paracetamol if needed).
+- For any serious or unclear issues, remind the user to see a doctor — but do it briefly and naturally.
+- Keep responses concise and under 4 sentences when possible.
+
+Tone:
+- Friendly, supportive, and calm.
+- No robotic warnings unless needed. Keep it real and human.
+- Use emojis like 😊 or 👍 occasionally to appear friendly.
+
+Important rules:
+- NEVER include text in parentheses in your responses.
+- NEVER include any meta-instructions in your responses.
+- NEVER include reminders about what you should do in future responses.
+- DO NOT include phrases like "We're here to help" or "I'm just an AI".
+- DO NOT include any text that instructs you what to do or how to behave.
+- DO NOT include any sentences that start with "If the user asks..." or "Remember..."
+- DO NOT include "(smile)" - instead, use actual emojis like 😊 or 👍 when appropriate.
+- DO NOT include numbered references like [1], [2], etc. in your responses.
+- DO NOT include any text that explains what your response is doing.
+- DO NOT include "user:" or "assistant:" prefixes in your responses.
+- DO NOT include hypothetical user questions in your responses.
+- DO NOT refuse to answer harmless non-medical questions like jokes or general knowledge.
+- Don't give exact dosages or diagnoses.
+- Be consistent in your responses regardless of the user's role."""
+
+# Global variables
+model = None
+tokenizer = None
+
+def load_model():
+    """Load FLAN-T5 Small model"""
+    global model, tokenizer
+    
+    try:
+        print(f"🏥 Loading FLAN-T5 Small for medical assistance...")
+        
+        # Load tokenizer
+        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+        
+        # Load model (FLAN-T5 is a seq2seq model)
+        model = AutoModelForSeq2SeqLM.from_pretrained(
+            MODEL_NAME,
+            torch_dtype=torch.float32,  # Use float32 for CPU
+            low_cpu_mem_usage=True
+        )
+        
+        print(f"✅ FLAN-T5 Small loaded successfully!")
+        return True
+        
+    except Exception as e:
+        print(f"❌ Failed to load model: {str(e)}")
+        return False
+
+def generate_response(prompt, max_tokens=None, temperature=None, top_p=None):
+    """Generate response using FLAN-T5"""
+    global model, tokenizer
+    
+    print(f"Starting generation for prompt: {prompt}")
+    if not prompt or not prompt.strip():
+        return "Please enter a question. 😊"
+    
+    if model is None or tokenizer is None:
+        return "❌ Model not loaded. Please wait for initialization."
+    
+    # Use defaults if not specified
+    max_tokens = max_tokens or MAX_NEW_TOKENS
+    temperature = temperature or TEMPERATURE
+    top_p = top_p or TOP_P
+    
+    try:
+        # Create instruction format for FLAN-T5
+        full_prompt = f"{MEDICAL_SYSTEM_PROMPT}\n\nQuestion: {prompt}\nAnswer:"
+        print(f"Full prompt: {full_prompt}")
+        
+        # Tokenize input
+        inputs = tokenizer(
+            full_prompt, 
+            return_tensors="pt", 
+            truncation=True, 
+            max_length=512,
+            padding=True
+        )
+        
+        # Generation parameters
+        generation_kwargs = {
+            "max_new_tokens": max_tokens,
+            "temperature": temperature,
+            "top_p": top_p,
+            "do_sample": True,
+            "repetition_penalty": 1.2,
+            "early_stopping": True
+        }
+        
+        print(f"Generating with kwargs: {generation_kwargs}")
+        
+        # Generate response
+        with torch.no_grad():
+            outputs = model.generate(**inputs, **generation_kwargs)
+        
+        # Decode response
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        
+        print(f"Generated response: {response}")
+        
+        # Clean up response
+        if not response or len(response.strip()) < 10:
+            response = "Sorry, I couldn't process that. Try again or see a doctor. 😊"
+        
+        # Memory cleanup
+        del inputs, outputs
+        gc.collect()
+        
+        return response
+        
+    except Exception as e:
+        print(f"��� Generation error: {str(e)}")
+        return f"I encountered a technical issue. Please try again! 😊"
+
+def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
+    """Main response function for Gradio ChatInterface"""
+    if not message or not message.strip():
+        return "Please enter a medical question or concern."
+    
+    try:
+        # Generate response
+        response = generate_response(
+            message.strip(),
+            max_tokens=int(max_tokens),
+            temperature=float(temperature),
+            top_p=float(top_p)
+        )
+        
+        return response
+        
+    except Exception as e:
+        print(f"System error: {str(e)}")
+        return f"⚠️ System temporarily unavailable. Please try again later or consult a healthcare professional directly."
+
+# Load model on startup
+print("🏥 Initializing FLAN-T5 Medical Chatbot...")
+model_loaded = load_model()
+
+if model_loaded:
+    print(f"✅ Ready! Using FLAN-T5 Small")
+else:
+    print("⚠️ WARNING: Model failed to load. The app will run but responses may be limited.")
+
+# Create Gradio interface
+demo = gr.ChatInterface(
+    respond,
+    title="🏥 Medical Assistant (FLAN-T5)",
+    description="A lightweight medical AI assistant powered by FLAN-T5 Small. Optimized for fast responses on CPU.",
+    additional_inputs=[
+        gr.Textbox(
+            value=MEDICAL_SYSTEM_PROMPT,
+            label="System Instructions",
+            lines=4,
+            interactive=False
+        ),
+        gr.Slider(
+            minimum=50, 
+            maximum=200, 
+            value=MAX_NEW_TOKENS, 
+            step=10, 
+            label="Max new tokens"
+        ),
+        gr.Slider(
+            minimum=0.1, 
+            maximum=1.0, 
+            value=TEMPERATURE, 
+            step=0.1, 
+            label="Temperature"
+        ),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=TOP_P,
+            step=0.05,
+            label="Top-p",
+        ),
+    ],
+    examples=[
+        ["What are the symptoms of diabetes?"],
+        ["How can I treat a headache?"],
+        ["What should I do for a fever?"],
+        ["Tell me about healthy eating"],
+        ["How to improve sleep quality?"]
+    ],
+    cache_examples=False,
+    theme=gr.themes.Soft(),
+    css=".gradio-container {max-width: 900px; margin: auto;}"
+)
+
+if __name__ == "__main__":
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=True,
+        show_error=True,
+        debug=True
+    )

config.py

@@ -2,37 +2,22 @@
 
 # Model configurations
 MODEL_CONFIGS = {
-    # Primary medical models (replace with actual MedLLaMA2 when available)
-    "medllama2": {
-        "name": "meta-llama/Llama-2-7b-chat-hf",  # Replace with actual MedLLaMA2 model ID
-        "description": "MedLLaMA2 7B medical language model"
-    },
-    
-    # Alternative medical models
     "meditron": {
         "name": "epfl-llm/meditron-7b",
         "description": "Meditron 7B medical language model"
     },
-    
-    "clinical_camel": {
-        "name": "wanglab/ClinicalCamel-70B",  # Note: This is very large, might not fit
-        "description": "Clinical Camel medical model"
-    },
-    
-    # Fallback models (smaller, more reliable)
     "dialogpt_medium": {
         "name": "microsoft/DialoGPT-medium",
         "description": "DialoGPT Medium (fallback)"
     },
-    
-    "dialogpt_small": {
-        "name": "microsoft/DialoGPT-small",
-        "description": "DialoGPT Small (lightweight fallback)"
+    "flan_t5_small": {
+        "name": "google/flan-t5-small",
+        "description": "FLAN-T5 Small (instruction-following fallback)"
     }
 }
 
-# Default model to use (switched to lighter model for better performance)
-DEFAULT_MODEL = "dialogpt_medium"
+# Default model to use - prioritize medical model
+DEFAULT_MODEL = "meditron"
 
 # Model loading settings (optimized for CPU)
 MODEL_SETTINGS = {