push

app.py

@@ -198,32 +198,61 @@ def process_query(user_input, history):
     yield 'Preparing your request 🛠️'
     
     # Step 1: Generate a search term based on the user query
-    stream_search = client.chat_completion(
+    messages = [
+    {
+        "role": "user",
+        "content": f"""Based on this chat history {history} the user's request '{user_input}', and this vector database {search_texts} - ignore it unless it relates to the user input, provide:
+- The number of searches needed to answer the query
+- The specific search terms required for each query, formatted as a Python object.
+
+The response should follow this exact format (example shown):
+{{ "number_of_searches_needed": 4,
+  "searches": ["First search term", "Second search term", "Third search term", "Fourth search term"]
+}}
+
+Return only the data in the specified format without additional text or explanation.
+IMPORTANT: NEVER ASSUME THE DATE IN ANY OF YOUR SEARCHES UNLESS THE USER GIVES YOU ONE, YOU DONT KNOW WHAT THE DATE IS BECAUSE OF YOUR KNOWLEDGE CUTOFF. ONLY SEARCH FOR RELEVANT THINGS BASED ON THE USER REQUEST FOR EXAMPLE IF THEY ASK ABOUT STOCK PRICE ONL;Y SEARCH FOR STOCK PRICE
+important: if the user input isnt complex use less searches in the intrest of time
+"""
+    }
+]
+
+    # Create streaming chat completion request
+    stream = client.chat.completions.create(
         model="Qwen/Qwen2.5-72B-Instruct",
-        messages=[{"role": "user", "content": f"Based on this chat history {history} the user's request '{user_input}', and this vector database {search_texts}, suggest a Google search term in a single line without specific dates; use 'this year', 'this month', etc. INCLUDE NOTHING IN YOUR RESPONSE EXCEPT THE RELEVANT SEARCH RESULT. EXAMPLE: USER: WHAT IS THE CURRENT PRICE OF COCA COLA STOCK. YOUR RESPONSE: WHAT IS THE CURRENT PRICE OF COCA COLA STOCK. here is their investor type."}],
-        max_tokens=400,
-        stream=True
+        messages=messages,
+        max_tokens=500,
+        stream=False
     )
 
-    # Collect the search term
-    search_query = ""
-    for chunk in stream_search:
-        content = chunk.choices[0].delta.content or ''
-        search_query += content
 
-    # Step 2: Perform the web search with the generated term
-    yield 'Searching the web for relevant information 🌐'
+    #take the response from the model and convert it into a list of searches
+    returned = stream.choices[0].message.content
+    dictionary_returned = json.loads(returned)
+    searches_needed = dictionary_returned['searches']
+    yield f'Searching the web for {searches_needed} 🌐'
+    completed_search = []
+
+
+    for value in searches_needed:
+      print(f'searching for {value}')
+      var = search(value)
+      completed_search.append(json.dumps(var))
+
+        
+    
+    
    
 
-    search_results = search(search_query)
+    
 
     # Format results as a JSON string for model input
-    search_results_str = json.dumps(search_results)
+    
     yield 'thinking...'
     # Step 3: Generate a response using the search results
     response = client.chat_completion(
         model="Qwen/Qwen2.5-72B-Instruct",
-        messages=[{"role": "user", "content": f"Using the search results: {search_results_str} and chat history {history}, this vector database on health checks {retrieved_texts} answer the user's query '{user_input}' in a concise, precise way,  using numerical data if available. GIVE DETAILED RESPONSES LIKE A STOCK ANALYST. THEY HAVE TOLD US HOW THEY INVEST SO WHEN YOU ANALYSE THINGS AND EXPLAIN, TAILOR IT TO THEM, AND REFERENCE THEIR TYPE IN YOUR ANSWERS - {investor_type_value}"}],
+        messages=[{"role": "user", "content": f"Using the search results: {completed_search} and chat history {history}, this vector database on health checks {retrieved_texts} answer the user's query '{user_input}' in an concise way,  using numerical data if available, follow the instructions from the vector database if they apply. tailor it to them - {investor_type_value}"}],
         max_tokens=3000,
         stream=True
     )
@@ -275,8 +304,7 @@ chatbot = gr.Chatbot(
     avatar_images=[None, BOT_AVATAR],
     show_copy_button=True,
     layout="panel",
-    height=700,
-   
+    height=700
 )
 
 with gr.Blocks(theme=theme) as demo: