Update app.py

app.py

@@ -89,12 +89,12 @@ embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
 # Initialize the models
 def initialize_phi_model():
     model = AutoModelForCausalLM.from_pretrained(
-        "microsoft/Phi-3.5-MoE-instruct", 
+        "microsoft/Phi-3.5-mini-instruct", 
         device_map="cuda", 
         torch_dtype="auto", 
         trust_remote_code=True, 
     )
-    tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3.5-MoE-instruct")
+    tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3.5-mini-instruct")
     return pipeline("text-generation", model=model, tokenizer=tokenizer)
 
 def initialize_gpt_model():
@@ -392,10 +392,8 @@ chain_neo4j = (
 # Short Prompt Template for Phi-3.5 Proprietary Model
 
 phi_short_template = f"""
-As an expert on Birmingham, Alabama, I will provide concise, accurate, and informative responses to your queries. Given the sunny weather today, {current_date}, feel free to ask me anything you need to know about the city.
-
+As an expert on Birmingham, Alabama, I will provide concise, accurate, and informative responses to your queries based on 256 token limit . Given the sunny weather today, {current_date}, feel free to ask me anything you need to know about the city.
 Provide only the direct answer to the question without any follow-up questions.
-
 {{context}}
 Question: {{question}}
 Answer:
@@ -456,7 +454,7 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
                 logging.debug(f"Phi-3.5 Prompt: {prompt}")
 
                 response = selected_model(prompt, **{
-                    "max_new_tokens": 128,  # Increased to handle longer responses
+                    "max_new_tokens": 256,  # Increased to handle longer responses
                     "return_full_text": False,
                     "temperature": 0.7,  # Adjusted to avoid cutting off
                     "do_sample": True,   # Allow sampling to increase response diversity