Updated with Content msg, product recommendation and history information4

app.py

@@ -8,6 +8,7 @@ import os
 import logging
 import re
 import torch
+import random  # For response variety
 
 # Set up logging
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
@@ -26,7 +27,6 @@ model_repo = "SyedHutter/blenderbot_model"
 model_subfolder = "blenderbot_model"
 model_dir = "/home/user/app/blenderbot_model"
 
-# Device setup
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 logger.info(f"Using device: {device}")
 
@@ -42,9 +42,9 @@ else:
     logger.info(f"Loading pre-existing model from {model_dir}.")
 tokenizer = BlenderbotTokenizer.from_pretrained(model_dir)
 model = BlenderbotForConditionalGeneration.from_pretrained(model_dir).to(device)
-model.eval()  # Set to evaluation mode for faster inference
+model.eval()
 
-# Static Context (shortened for efficiency)
+# Static Context
 context_msg = "I am Hutter, your shopping guide for Hutter Products GmbH, here to help you find sustainable products."
 
 # spaCy Setup
@@ -74,11 +74,11 @@ def detect_intent(text: str) -> str:
         return "recommend_product"
     elif any(token.text in ["company", "who", "do"] for token in doc):
         return "company_info"
-    elif "name" in text_lower:
+    elif "name" in text_lower or "yourself" in text_lower or "you" in doc and "about" in doc:
         return "ask_name"
     elif re.search(r"\d+\s*[\+\-\*/]\s*\d+", text_lower):
         return "math_query"
-    return "recommend_product"
+    return "chat"  # New fallback for general conversation
 
 def search_products_by_keywords(keywords: List[str]) -> List[Dict[str, Any]]:
     if not keywords:
@@ -101,16 +101,22 @@ def get_product_context(products: List[Dict]) -> str:
     product_str = "Products: " + ", ".join([f"'{p['name']}' - {p['description']}" for p in products[:2]])
     return product_str
 
-def format_response(response: str, products: List[Dict], intent: str, input_text: str) -> str:
+def format_response(response: str, products: List[Dict], intent: str, input_text: str, history: List[str]) -> str:
+    no_product_prompts = [
+        "I’d love to recommend something! What are you looking for in our sustainable catalog?",
+        "Our sustainable catalog has lots to offer—what catches your interest?",
+        "Tell me what you’re after, and I’ll find something great from our eco-friendly range!"
+    ]
+    
     if intent == "recommend_product":
         if not products:
-            return "I’d love to recommend something! What are you looking for in our sustainable catalog?"
+            return random.choice(no_product_prompts)
         product = products[0]
         return f"Check out our '{product['name']}'—it’s {product['description'].lower()}. Want more options?"
     elif intent == "company_info":
         return "Hutter Products GmbH offers sustainable products like recycled textiles and ocean plastic goods."
     elif intent == "ask_name":
-        return "I’m Hutter, your shopping guide for Hutter Products GmbH. How can I help?"
+        return "I’m Hutter, your shopping guide for Hutter Products GmbH. I’m here to help you find eco-friendly products—how can I assist?"
     elif intent == "math_query":
         match = re.search(r"(\d+)\s*([\+\-\*/])\s*(\d+)", input_text.lower())
         if match:
@@ -120,6 +126,11 @@ def format_response(response: str, products: List[Dict], intent: str, input_text
             elif op == "*": return f"{num1} * {num2} = {num1 * num2}. Explore our products?"
             elif op == "/": return f"{num1} / {num2} = {num1 / num2}." if num2 != 0 else "Can’t divide by zero! Try our products?"
         return "I can do math—try '2 + 2'. What else can I help with?"
+    elif intent == "chat":
+        # Use BlenderBot’s response if appropriate, else nudge toward shopping
+        if "yes" in input_text.lower() and history and "hat" in history[-1].lower():
+            return "Great! Besides the Bucket Hat, we have other sustainable items—want to hear more?"
+        return f"{response} How can I assist with our sustainable products today?" if response else "I’m here to help—anything on your mind?"
     if products:
         product = products[0]
         return f"{response} Also, check out '{product['name']}'—it’s {product['description'].lower()}."
@@ -135,7 +146,7 @@ async def process_prompt(request: PromptRequest):
     try:
         logger.info(f"Processing request: {request.input_text}")
         input_text = request.input_text
-        history = request.conversation_history[-1:] if request.conversation_history else []  # Limit to last message
+        history = request.conversation_history[-1:] if request.conversation_history else []
 
         intent = detect_intent(input_text)
         keywords = extract_keywords(input_text)
@@ -150,17 +161,17 @@ async def process_prompt(request: PromptRequest):
         logger.info(f"Full input to model: {full_input}")
 
         logger.info("Tokenizing input...")
-        inputs = tokenizer(full_input, return_tensors="pt", truncation=True, max_length=64).to(device)  # Reduced max_length
+        inputs = tokenizer(full_input, return_tensors="pt", truncation=True, max_length=64).to(device)
         logger.info("Input tokenized successfully.")
 
         logger.info("Generating model response...")
-        with torch.no_grad():  # Disable gradient computation
+        with torch.no_grad():
             outputs = model.generate(
                 **inputs,
-                max_new_tokens=30,  # Limit new tokens for speed
-                do_sample=True,     # Faster sampling over beam search
-                top_p=0.9,          # Nucleus sampling
-                temperature=0.7,    # Controlled randomness
+                max_new_tokens=30,
+                do_sample=True,
+                top_p=0.95,  # Slightly higher for more variety
+                temperature=0.8,  # Slightly higher for creativity
                 no_repeat_ngram_size=2
             )
         logger.info("Model generation complete.")
@@ -169,7 +180,7 @@ async def process_prompt(request: PromptRequest):
         response = tokenizer.decode(outputs[0], skip_special_tokens=True)
         logger.info(f"Model response: {response}")
 
-        enhanced_response = format_response(response, products, intent, input_text)
+        enhanced_response = format_response(response, products, intent, input_text, request.conversation_history)
         qa_response = {
             "question": input_text,
             "answer": enhanced_response,