Update app.py

app.py

@@ -25,15 +25,18 @@ WRITE_ACCESS_TOKEN = st.secrets["Llama_3_1"]
 def get_rag_answer(query, top_results):
     """
     Constructs a prompt from the query and the page contexts of the top results,
-    then sends it to the dedicated endpoint and returns the generated answer.
+    then sends it to the dedicated endpoint and returns only the generated answer.
     """
     # Combine the context from the top results (you may adjust the separator as needed)
     context = "\n\n".join([res.payload["page_content"] for res in top_results])
     
-    # Create a prompt: you can refine the instructions to better suit your needs.
+    # Create a prompt that instructs the model to output only the answer.
     prompt = (
-        f"Looking at the Top 15 Projects, the answer your question is.\n\n"
-        f"Answer:"
+        "Using the following context, answer the question concisely. "
+        "Only output the final answer below, without repeating the context or question.\n\n"
+        f"Context:\n{context}\n\n"
+        f"Question: {query}\n\n"
+        "Answer:"
     )
     
     headers = {"Authorization": f"Bearer {WRITE_ACCESS_TOKEN}"}
@@ -47,9 +50,11 @@ def get_rag_answer(query, top_results):
     response = requests.post(DEDICATED_ENDPOINT, headers=headers, json=payload)
     if response.status_code == 200:
         result = response.json()
-        # Depending on the endpoint's response structure, adjust how you extract the generated text.
         answer = result[0]["generated_text"]
-        return answer.strip()
+        # If the model returns the full prompt, split and extract only the portion after "Answer:"
+        if "Answer:" in answer:
+            answer = answer.split("Answer:")[-1].strip()
+        return answer
     else:
         return f"Error in generating answer: {response.text}"