Update app.py

app.py

@@ -250,7 +250,7 @@ def respond(message, history, model, temperature, num_calls, use_web_search):
 
 logging.basicConfig(level=logging.DEBUG)
 
-def get_response_from_cloudflare(prompt, num_calls=3, temperature=0.2):
+def get_response_from_cloudflare(prompt, context, query, num_calls=3, temperature=0.2):
     headers = {
         "Authorization": f"Bearer {API_TOKEN}",
         "Content-Type": "application/json"
@@ -259,7 +259,7 @@ def get_response_from_cloudflare(prompt, num_calls=3, temperature=0.2):
 
     inputs = [
         {"role": "system", "content": "You are a friendly assistant that helps answer questions based on provided context."},
-        {"role": "user", "content": prompt}
+        {"role": "user", "content": f"Context: {context}\n\nQuestion: {query}"}
     ]
 
     payload = {
@@ -343,7 +343,7 @@ Write a detailed and complete response that answers the following user question:
 
     if model == "@cf/meta/llama-3.1-8b-instruct":
         # Use Cloudflare API with the retrieved context
-        for response in get_response_from_cloudflare(prompt, num_calls, temperature):
+        for response in get_response_from_cloudflare(prompt, context_str, query, num_calls, temperature):
             yield response
     else:
         # Use Hugging Face API