Update app.py

app.py

@@ -1,10 +1,9 @@
 import logging
 import os
-import time  # Added for timing logs
+import time
 import torch
 import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
-import difflib
 import json
 
 # Set up logging
@@ -54,6 +53,46 @@ response_cache = {
         "Invest what you can afford after expenses and an emergency fund. Start with $100-$500 monthly "
         "in ETFs like VOO using dollar-cost averaging. Consult a financial planner."
     ),
+    "how to start investing": (
+        "Here’s how to start investing:\n"
+        "1. Educate yourself using resources like Investopedia.\n"
+        "2. Open a brokerage account with a platform like Fidelity.\n"
+        "3. Deposit an initial amount, such as $100, after building an emergency fund.\n"
+        "4. Choose a low-cost ETF like VOO.\n"
+        "5. Invest regularly using dollar-cost averaging.\n"
+        "Consult a financial planner for personalized advice."
+    ),
+    "best brokerage accounts": (
+        "The best brokerage accounts for beginners include Fidelity, Vanguard, Charles Schwab, and Robinhood. "
+        "They offer low fees, no minimums, and user-friendly platforms."
+    ),
+    "investing for beginners": (
+        "Here’s a beginner’s guide to investing:\n"
+        "1. Learn the basics from Investopedia or books like 'The Intelligent Investor.'\n"
+        "2. Set clear investment goals and assess your risk tolerance.\n"
+        "3. Open a brokerage account with a platform like Fidelity or Robinhood.\n"
+        "4. Start with low-cost ETFs like VOO or index funds.\n"
+        "5. Invest regularly using dollar-cost averaging.\n"
+        "6. Monitor your investments quarterly.\n"
+        "Consult a financial planner for tailored advice."
+    ),
+    "steps to start investing": (
+        "Here are the steps to start investing:\n"
+        "1. Educate yourself on investing basics.\n"
+        "2. Open a brokerage account with a beginner-friendly platform.\n"
+        "3. Deposit an initial amount you can afford.\n"
+        "4. Choose a diversified investment like an ETF.\n"
+        "5. Invest consistently over time.\n"
+        "Consult a financial planner for more guidance."
+    ),
+    "recommended etfs": (
+        "Recommended ETFs for beginners include VOO (tracks S&P 500), QQQ (tech-focused), and VT (global market exposure). "
+        "They offer diversification and low fees."
+    ),
+    "how much to invest": (
+        "The amount to invest depends on your financial situation. Start with what you can afford after covering expenses and an emergency fund. "
+        "A common starting point is $100-$500 monthly in low-cost ETFs like VOO. Consult a financial planner for personalized advice."
+    ),
 }
 
 # Load persistent cache
@@ -88,7 +127,7 @@ prompt_prefix = (
     "Avoid vague or unrelated topics. Use a numbered list format where appropriate and explain each step.\n\n"
     "Example 1:\n"
     "Q: How can I start investing with $100 a month?\n"
-    "A: Here’s a step-by point-by-step guide:\n"
+    "A: Here’s a step-by-step guide:\n"
     "1. Open a brokerage account with a platform like Fidelity or Robinhood. They offer low fees and no minimums.\n"
     "2. Deposit your $100 monthly. You can set up automatic transfers.\n"
     "3. Choose a low-cost ETF like VOO, which tracks the S&P 500.\n"
@@ -101,23 +140,20 @@ prompt_prefix = (
     "Q: "
 )
 
-# Fuzzy matching for cache
-def get_closest_cache_key(message, cache_keys, threshold=0.7):
-    matches = difflib.get_close_matches(message, cache_keys, n=1, cutoff=threshold)
-    return matches[0] if matches else None
-
-# Define chat function with optimized generation parameters
+# Define chat function with substring matching and reduced max_new_tokens
 def chat_with_model(user_input, history=None):
     try:
-        start_time = time.time()  # Start timing
+        start_time = time.time()
         logger.info(f"Processing user input: {user_input}")
         
-        cache_key = user_input.lower().strip()
-        cache_keys = list(response_cache.keys())
-        closest_key = cache_key if cache_key in response_cache else get_closest_cache_key(cache_key, cache_keys)
-        if closest_key:
-            logger.info(f"Cache hit for: {closest_key}")
-            response = response_cache[closest_key]
+        user_input_lower = user_input.lower().strip()
+        
+        # Substring matching for cache
+        matching_keys = [key for key in response_cache if key in user_input_lower]
+        if matching_keys:
+            longest_key = max(matching_keys, key=len)
+            logger.info(f"Cache hit for: {longest_key}")
+            response = response_cache[longest_key]
             logger.info(f"Chatbot response: {response}")
             history = history or []
             history.append({"role": "user", "content": user_input})
@@ -141,12 +177,12 @@ def chat_with_model(user_input, history=None):
         inputs = tokenizer(full_prompt, return_tensors="pt", truncation=True, max_length=512).to(device)
         
         with torch.inference_mode():
-            gen_start_time = time.time()  # Start generation timing
+            gen_start_time = time.time()
             outputs = model.generate(
                 **inputs,
-                max_new_tokens=75,  # Reduced for faster generation
+                max_new_tokens=50,  # Reduced for faster generation
                 min_length=20,
-                do_sample=False,  # Use greedy decoding for speed
+                do_sample=False,  # Greedy decoding for speed
                 repetition_penalty=1.2,
                 pad_token_id=tokenizer.eos_token_id
             )
@@ -157,7 +193,8 @@ def chat_with_model(user_input, history=None):
         response = response[len(full_prompt):].strip() if response.startswith(full_prompt) else response
         logger.info(f"Chatbot response: {response}")
         
-        response_cache[cache_key] = response
+        # Update cache with exact user input as key
+        response_cache[user_input_lower] = response
         logger.info("Cache miss, added to in-memory cache")
         
         history = history or []