Update app.py

app.py

@@ -3,7 +3,7 @@ import numpy as np
 import re
 from datetime import datetime, timedelta
 import difflib
-from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 import yfinance as yf
 from functools import lru_cache
@@ -36,7 +36,7 @@ def parse_period(query):
             return timedelta(weeks=num)
         elif unit == 'day':
             return timedelta(days=num)
-    return timedelta(days=365)  # Default to 1 year
+    return timedelta(days=365) # Default to 1 year
 
 def find_closest_symbol(input_symbol):
     input_symbol = input_symbol.upper()
@@ -55,21 +55,16 @@ def calculate_growth_rate(start_date, end_date, symbol):
     if years == 0:
         return 0
     cagr = (1 + total_return) ** (1 / years) - 1
-    return cagr * 100  # As percentage
+    return cagr * 100 # As percentage
 
 def calculate_investment(principal, years, annual_return=0.07):
     return principal * (1 + annual_return) ** years
 
-# Load SmolLM-135M-Instruct with 8-bit quantization
+# Load SmolLM-135M-Instruct without quantization
 model_name = "HuggingFaceTB/SmolLM-135M-Instruct"
-quantization_config = BitsAndBytesConfig(
-    load_in_8bit=True,  # Switch to 8-bit for faster inference
-    bnb_8bit_compute_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-)
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
-    quantization_config=quantization_config,
     device_map="auto",
 )
 
@@ -78,7 +73,6 @@ def generate_response(user_query, enable_thinking=False):
     stock_keywords = ['stock', 'growth', 'investment', 'price', 'return', 'cagr']
     is_stock_query = any(keyword in user_query.lower() for keyword in stock_keywords)
     summary = ""
-
     if is_stock_query:
         # Parse query for symbol and period
         symbol_match = re.search(r'\b([A-Z]{1,5})\b', user_query.upper())
@@ -87,22 +81,19 @@ def generate_response(user_query, enable_thinking=False):
             period = parse_period(user_query)
             end_date = datetime.now()
             start_date = end_date - period
-
             # Calculate growth rate
             growth_rate = calculate_growth_rate(start_date, end_date, symbol)
             if growth_rate is not None:
                 summary = f"The CAGR for {symbol} over the period is {growth_rate:.2f}%."
             else:
                 summary = f"No data available for {symbol} in the specified period."
-
             # Handle investment projection
             investment_match = re.search(r'\$(\d+)', user_query)
             if investment_match:
                 principal = float(investment_match.group(1))
                 years = period.days / 365.25
                 projected = calculate_investment(principal, years)
-                summary += f" Projecting ${principal} at 7% return over {years:.1f} years: ${projected:.2f}."
-
+                summary += f" Projecting $ {principal} at 7% return over {years: .1f} years: $ {projected: .2f}."
     # Prepare prompt
     system_prompt = (
         "You are FinChat, a knowledgeable financial advisor. Always respond in a friendly, professional manner. "
@@ -112,44 +103,37 @@ def generate_response(user_query, enable_thinking=False):
     )
     messages = [
         {"role": "system", "content": system_prompt},
-        {"role": "user", "content": f"{summary} {user_query}" if summary else user_query}
+        {"role": "user", "content": f" {summary} {user_query}" if summary else user_query}
     ]
-
     text = tokenizer.apply_chat_template(
         messages,
         tokenize=False,
         add_generation_prompt=True,
         enable_thinking=enable_thinking
     )
-
     model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
     generated_ids = model.generate(
         **model_inputs,
-        max_new_tokens=30,  # Reduced for speed
+        max_new_tokens=30, # Reduced for speed
         temperature=0.6,
         top_p=0.95,
         repetition_penalty=1.0,
         do_sample=False,
-        early_stopping=True  # Stop early for efficiency
+        early_stopping=True # Stop early for efficiency
     )
-
     output_ids = generated_ids[0][len(model_inputs.input_ids[0]):]
     response = tokenizer.decode(output_ids, skip_special_tokens=True)
     return response.strip()
-
 # Gradio interface
 def chat(user_input, history):
     response = generate_response(user_input)
     history.append((user_input, response))
     return history, ""
-
 with gr.Blocks() as demo:
     gr.Markdown("# FinChat: AI-Powered Financial Advisor")
     chatbot = gr.Chatbot()
     msg = gr.Textbox(placeholder="Ask about stocks, investments, etc.")
     clear = gr.Button("Clear")
-
     msg.submit(chat, [msg, chatbot], [chatbot, msg])
     clear.click(lambda: None, None, chatbot, queue=False)
-
 demo.launch()