Update app.py

app.py

@@ -12,11 +12,21 @@ logger = logging.getLogger(__name__)
 device = torch.device("cpu")
 logger.info(f"Using device: {device}")
 
-# Response cache with consolidated and new entries
+# Response cache with new entry for step-by-step advice
 response_cache = {
     "hi": "Hello! I'm your financial advisor. How can I help with investing?",
     "hello": "Hello! I'm your financial advisor. How can I help with investing?",
     "hey": "Hi there! Ready to discuss investment goals?",
+    "hi, give me step-by-step investing advice": (
+        "Here’s a step-by-step guide to start investing:\n"
+        "1. **Open a Brokerage Account**: If you’re 18 or older, sign up with a platform like Fidelity, Vanguard, or Robinhood.\n"
+        "2. **Deposit Funds**: Add an initial amount you can afford, such as $100, after securing an emergency fund.\n"
+        "3. **Research and Buy**: Choose a stock, ETF (e.g., VOO for S&P 500), or index fund based on research from Yahoo Finance or Morningstar.\n"
+        "4. **Monitor Investments**: Check your portfolio regularly and enable dividend reinvesting for compounding returns.\n"
+        "5. **Use Dollar-Cost Averaging**: Invest a fixed amount (e.g., $100 monthly) consistently to reduce market timing risks.\n"
+        "6. **Diversify**: Spread investments across sectors to manage risk.\n"
+        "Consult a certified financial planner for personalized advice."
+    ),
     "hi, pretend you are a financial advisor. now tell me how can i start investing in stock market?": (
         "Here’s a guide to start investing in the stock market:\n"
         "1. **Learn**: Use Investopedia or 'The Intelligent Investor' by Benjamin Graham.\n"
@@ -73,13 +83,17 @@ except Exception as e:
     logger.error(f"Error loading model/tokenizer: {e}")
     raise
 
-# Pre-tokenize prompt prefix with few-shot example
+# Pre-tokenize prompt prefix with improved few-shot example
 prompt_prefix = (
-    "You are a financial advisor. Provide concise, actionable advice in a numbered list when asked for ideas or steps. "
-    "If no specific recommendations, suggest alternatives. Avoid vague statements.\n\n"
+    "You are a financial advisor. Provide concise, actionable advice in a numbered list for step-by-step or idea prompts. "
+    "Avoid repetition and vague statements. Use varied, specific steps.\n\n"
     "Example:\n"
-    "Q: Give me some investing tips\n"
-    "A: 1. Open a brokerage account (e.g., Fidelity).\n2. Start with $100 in an ETF like VOO.\n3. Research investments on Yahoo Finance.\n4. Invest regularly using dollar-cost averaging.\n\n"
+    "Q: Give me step-by-step investing advice\n"
+    "A: 1. Open a brokerage account with Fidelity or Vanguard if 18 or older.\n"
+    "2. Deposit an affordable amount, like $100, after building an emergency fund.\n"
+    "3. Research and buy an ETF like VOO using Yahoo Finance data.\n"
+    "4. Check investments monthly and enable dividend reinvesting.\n"
+    "5. Invest regularly with dollar-cost averaging to reduce risk.\n\n"
     "Q: "
 )
 prefix_tokens = tokenizer(prompt_prefix, return_tensors="pt", truncation=True, max_length=512).to(device)
@@ -107,8 +121,12 @@ def chat_with_model(message, history=None):
         with torch.no_grad():
             outputs = model.generate(
                 **inputs,
-                max_new_tokens=30,  # Increased for detailed lists
-                do_sample=False,  # Greedy decoding for speed
+                max_new_tokens=100,  # Increased for detailed lists
+                min_length=20,  # Encourage substantive responses
+                do_sample=True,  # Sampling to avoid repetition
+                temperature=0.7,  # Balanced diversity
+                top_p=0.9,  # Nucleus sampling
+                no_repeat_ngram_size=2,  # Prevent repetitive phrases
                 pad_token_id=tokenizer.eos_token_id
             )
         response = tokenizer.decode(outputs[0], skip_special_tokens=True)
@@ -126,8 +144,8 @@ interface = gr.ChatInterface(
     description="Ask about investing! Powered by Meta AI's OPT-350m. Fast, detailed answers.",
     examples=[
         "Hi",
+        "Hi, give me step-by-step investing advice",
         "Give me few investing idea",
-        "Hi, pretend you are a financial advisor. Now tell me how can I start investing in stock market?",
         "Do you have a list of companies you recommend?",
         "What's the difference between stocks and bonds?",
         "How much should I invest?"
@@ -135,12 +153,7 @@ interface = gr.ChatInterface(
 )
 
 # Launch interface (conditional for Spaces)
-if __name__ == "__main__" and not os.getenv("HF_SPACE"):
+if __name__ == "__main__ and not os.getenv("HF_SPACE"):
     logger.info("Launching Gradio interface locally")
     try:
-        interface.launch(share=False, debug=True)
-    except Exception as e:
-        logger.error(f"Error launching interface: {e}")
-        raise
-else:
-    logger.info("Running in Hugging Face Spaces, interface defined but not launched")
+        interface.launch(