Update app.py

app.py

@@ -471,6 +471,75 @@ def get_response_from_excel(query, model, context, num_calls=3, temperature=0.2)
 
         logging.info("Finished generating response for Excel data")
 
+def get_response_from_llama(query, model, selected_docs, file_type, num_calls=1, temperature=0.2):
+    logging.info(f"Getting response from Llama using model: {model}")
+    
+    # Initialize the Hugging Face client
+    client = InferenceClient(model, token=huggingface_token)
+    
+    if file_type == "excel":
+        # Excel functionality
+        system_instruction = """You are a highly specialized Python programmer with deep expertise in data analysis and visualization using Excel spreadsheets. 
+        Your primary goal is to generate accurate and efficient Python code to perform calculations or create visualizations based on the user's requests. 
+        Strictly use the data provided to write code that identifies key metrics, trends, and significant details relevant to the query. 
+        Do not make assumptions or include any information that is not explicitly supported by the dataset. 
+        If the user requests a calculation, provide the appropriate Python code to execute it, and if a visualization is needed, generate code using the matplotlib library to create the chart."""
+        
+        # Get the context from selected Excel documents
+        embed = get_embeddings()
+        office_database = FAISS.load_local("office_faiss_database", embed, allow_dangerous_deserialization=True)
+        retriever = office_database.as_retriever(search_kwargs={"k": 10})
+        relevant_docs = retriever.get_relevant_documents(query)
+        context = "\n".join([doc.page_content for doc in relevant_docs if doc.metadata["source"] in selected_docs])
+        
+        messages = [
+            {"role": "system", "content": system_instruction},
+            {"role": "user", "content": f"Based on the following data extracted from Excel spreadsheets:\n{context}\n\nPlease provide the Python code needed to execute the following task: '{query}'. Ensure that the code is derived directly from the dataset. If a chart is requested, use the matplotlib library to generate the appropriate visualization."}
+        ]
+        
+    elif file_type == "pdf":
+        # PDF functionality
+        embed = get_embeddings()
+        pdf_database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
+        
+        retriever = pdf_database.as_retriever(search_kwargs={"k": 10})
+        relevant_docs = retriever.get_relevant_documents(query)
+        
+        context_str = "\n".join([doc.page_content for doc in relevant_docs if doc.metadata["source"] in selected_docs])
+    
+        system_instruction = """You are a highly specialized financial analyst assistant with expertise in analyzing and summarizing financial documents. 
+        Your goal is to provide accurate, detailed, and precise summaries based on the context provided. 
+        Avoid making assumptions or adding information that is not explicitly supported by the context from the PDF documents."""
+
+        messages = [
+            {"role": "system", "content": system_instruction},
+            {"role": "user", "content": f"Using the following context from the PDF documents:\n{context_str}\n\nPlease generate a step-by-step reasoning before arriving at a comprehensive and accurate summary addressing the following question: '{query}'. Ensure your response is strictly based on the provided context, highlighting key metrics, trends, and significant details relevant to the query. Avoid any speculative or unverified information."}
+        ]
+
+    else:
+        raise ValueError("Invalid file type. Use 'excel' or 'pdf'.")
+
+    full_response = ""
+    for _ in range(num_calls):
+        try:
+            # Generate content with streaming enabled
+            for message in client.chat_completion(
+                messages=messages,
+                max_new_tokens=20000,
+                temperature=temperature,
+                stream=True,
+                top_p=0.8,
+            ):
+                if message.choices and message.choices[0].delta and message.choices[0].delta.content:
+                    chunk = message.choices[0].delta.content
+                    full_response += chunk
+                    yield full_response  # Yield the accumulated response so far
+        except Exception as e:
+            yield f"An error occurred with the Llama model: {str(e)}. Please try again."
+
+    if not full_response:
+        yield "No response generated from the Llama model."
+
 # Modify the existing respond function to handle both PDF and web search
 def respond(message, history, model, temperature, num_calls, use_web_search, selected_docs):
     logging.info(f"User Query: {message}")
@@ -500,7 +569,6 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
             yield final_summary
         else:
             yield "Unable to generate a response. Please try a different query."
-
     else:
         try:
             embed = get_embeddings()
@@ -548,6 +616,9 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
                 if model == "gemini-1.5-flash":
                     for chunk in get_response_from_gemini(message, model, selected_docs, file_type, num_calls, temperature):
                         yield chunk
+                elif "llama" in model.lower():
+                    for chunk in get_response_from_llama(message, model, selected_docs, file_type, num_calls, temperature):
+                        yield chunk
                 else:
                     for response in get_response_from_excel(message, model, excel_context, num_calls, temperature):
                         yield response
@@ -561,6 +632,9 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
                 elif model == "@cf/meta/llama-3.1-8b-instruct":
                     for response in get_response_from_cloudflare(prompt="", context=other_context, query=message, num_calls=num_calls, temperature=temperature, search_type="document"):
                         yield response
+                elif "llama" in model.lower():
+                    for chunk in get_response_from_llama(message, model, selected_docs, file_type, num_calls, temperature):
+                        yield chunk
                 else:
                     for response in get_response_from_pdf(message, model, selected_docs, num_calls, temperature):
                         yield response