Updated with Content msg, product recommendation and history information3

app.py

@@ -7,8 +7,9 @@ import spacy
 import os
 import logging
 import re
+import torch
 
-# Set up logging with detailed output
+# Set up logging
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
 logger = logging.getLogger(__name__)
 
@@ -25,6 +26,10 @@ 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}")
+
 if not os.path.exists(model_dir):
     logger.info(f"Downloading {model_repo}/{model_subfolder} to {model_dir}...")
     tokenizer = BlenderbotTokenizer.from_pretrained(model_repo, subfolder=model_subfolder)
@@ -35,12 +40,12 @@ if not os.path.exists(model_dir):
     logger.info("Model download complete.")
 else:
     logger.info(f"Loading pre-existing model from {model_dir}.")
-
 tokenizer = BlenderbotTokenizer.from_pretrained(model_dir)
-model = BlenderbotForConditionalGeneration.from_pretrained(model_dir)
+model = BlenderbotForConditionalGeneration.from_pretrained(model_dir).to(device)
+model.eval()  # Set to evaluation mode for faster inference
 
-# Static Context
-context_msg = "I am Hutter, your shopping guide for Hutter Products GmbH. I’m here to help you explore our innovative and sustainable product catalog, featuring eco-friendly items like recycled textiles and ocean plastic goods. Let me assist you in finding the perfect sustainable solution!"
+# Static Context (shortened for efficiency)
+context_msg = "I am Hutter, your shopping guide for Hutter Products GmbH, here to help you find sustainable products."
 
 # spaCy Setup
 spacy_model_path = "/home/user/app/en_core_web_sm-3.8.0"
@@ -65,7 +70,6 @@ def extract_keywords(text: str) -> List[str]:
 def detect_intent(text: str) -> str:
     doc = nlp(text.lower())
     text_lower = text.lower()
-    # General product-related intent based on shopping context
     if any(token.text in ["buy", "shop", "find", "recommend", "product", "products", "item", "store", "catalog"] for token in doc) or "what" in text_lower.split()[:2]:
         return "recommend_product"
     elif any(token.text in ["company", "who", "do"] for token in doc):
@@ -74,11 +78,10 @@ def detect_intent(text: str) -> str:
         return "ask_name"
     elif re.search(r"\d+\s*[\+\-\*/]\s*\d+", text_lower):
         return "math_query"
-    return "recommend_product"  # Default to product focus for scalability
+    return "recommend_product"
 
 def search_products_by_keywords(keywords: List[str]) -> List[Dict[str, Any]]:
     if not keywords:
-        logger.info("No keywords provided, returning empty product list.")
         return []
     query = {"$or": [{"name": {"$regex": f"\\b{keyword}\\b", "$options": "i"}} for keyword in keywords]}
     matched_products = [
@@ -95,39 +98,32 @@ def search_products_by_keywords(keywords: List[str]) -> List[Dict[str, Any]]:
 def get_product_context(products: List[Dict]) -> str:
     if not products:
         return ""
-    product_str = "Available products: "
-    product_str += ", ".join([f"'{p['name']}' - {p['description']}" for p in products[:2]])
+    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:
-    # Handle product recommendation intent
     if intent == "recommend_product":
         if not products:
-            return "I’d love to recommend something from our sustainable catalog! Could you tell me more about what you’re looking for?"
+            return "I’d love to recommend something! What are you looking for in our sustainable catalog?"
         product = products[0]
-        return f"Check out our '{product['name']}'—it’s {product['description'].lower()}. Want to explore more options?"
+        return f"Check out our '{product['name']}'—it’s {product['description'].lower()}. Want more options?"
     elif intent == "company_info":
-        return "Hutter Products GmbH specializes in sustainable product design and production, offering eco-friendly items like recycled textiles and ocean plastic goods."
+        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 assist you today?"
+        return "I’m Hutter, your shopping guide for Hutter Products GmbH. How can I help?"
     elif intent == "math_query":
         match = re.search(r"(\d+)\s*([\+\-\*/])\s*(\d+)", input_text.lower())
         if match:
             num1, op, num2 = int(match.group(1)), match.group(2), int(match.group(3))
-            if op == "+":
-                return f"{num1} plus {num2} is {num1 + num2}. Need help with shopping too?"
-            elif op == "-":
-                return f"{num1} minus {num2} is {num1 - num2}. Anything else I can assist with?"
-            elif op == "*":
-                return f"{num1} times {num2} is {num1 * num2}. Want to explore our products?"
-            elif op == "/":
-                return f"{num1} divided by {num2} is {num1 / num2}." if num2 != 0 else "Can’t divide by zero! How about some sustainable products instead?"
-        return "I can do simple math—try '2 + 2'. What else can I help you with?"
-    # Fallback with product nudge if available
+            if op == "+": return f"{num1} + {num2} = {num1 + num2}. Need shopping help?"
+            elif op == "-": return f"{num1} - {num2} = {num1 - num2}. Anything else?"
+            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?"
     if products:
         product = products[0]
-        return f"{response} By the way, how about our '{product['name']}'? It’s {product['description'].lower()}."
-    return response if response else "How can I assist you with our sustainable products today?"
+        return f"{response} Also, check out '{product['name']}'—it’s {product['description'].lower()}."
+    return response if response else "How can I assist with our sustainable products?"
 
 # Endpoints
 @app.get("/")
@@ -139,7 +135,7 @@ async def process_prompt(request: PromptRequest):
     try:
         logger.info(f"Processing request: {request.input_text}")
         input_text = request.input_text
-        history = request.conversation_history[-3:] if request.conversation_history else []
+        history = request.conversation_history[-1:] if request.conversation_history else []  # Limit to last message
 
         intent = detect_intent(input_text)
         keywords = extract_keywords(input_text)
@@ -150,15 +146,23 @@ async def process_prompt(request: PromptRequest):
         logger.info(f"Products matched: {len(products)}")
 
         history_str = " || ".join(history)
-        full_input = f"{context_msg} || {history_str} || {product_context} || {input_text}" if (history or product_context) else f"{context_msg} || {input_text}"
+        full_input = f"{context_msg} || {product_context} || {input_text}" if product_context else f"{context_msg} || {input_text}"
         logger.info(f"Full input to model: {full_input}")
 
         logger.info("Tokenizing input...")
-        inputs = tokenizer(full_input, return_tensors="pt", truncation=True, max_length=512)
+        inputs = tokenizer(full_input, return_tensors="pt", truncation=True, max_length=64).to(device)  # Reduced max_length
         logger.info("Input tokenized successfully.")
 
         logger.info("Generating model response...")
-        outputs = model.generate(**inputs, max_length=50, num_beams=4, length_penalty=1.0, no_repeat_ngram_size=2)
+        with torch.no_grad():  # Disable gradient computation
+            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
+                no_repeat_ngram_size=2
+            )
         logger.info("Model generation complete.")
 
         logger.info("Decoding model output...")
@@ -180,7 +184,7 @@ async def process_prompt(request: PromptRequest):
         }
     except Exception as e:
         logger.error(f"Error processing request: {str(e)}", exc_info=True)
-        raise HTTPException(status_code=500, detail=f"Oops, something went wrong: {str(e)}. Try again!")
+        raise HTTPException(status_code=500, detail=f"Oops, something went wrong: {str(e)}")
 
 @app.on_event("startup")
 async def startup_event():