Update app.py

app.py

@@ -340,7 +340,7 @@ class RAGIndex:
 
         output_ids = self.qa_model.generate(
             **inputs,
-            max_new_tokens=256,
+            max_new_tokens=128,
             do_sample=False,
         )
 
@@ -381,10 +381,12 @@ class RAGIndex:
 
         for ctx, source, score in contexts:
             used_sources.add(source)
-            combined_context.append(f"[Source: {source}]\n{ctx}")
+            # Only include the pure text as context
+            combined_context.append(ctx)
 
         combined_text = "\n\n".join(combined_context)
 
+
         # Limit context length to keep it manageable
         max_context_chars = 4000
         if len(combined_text) > max_context_chars:
@@ -392,12 +394,14 @@ class RAGIndex:
 
         # Prompt for the generative model
         prompt = (
-            "You are an AI assistant that answers questions using only the provided context. "
-            "If the answer cannot be found in the context, reply exactly with: "
+            "You are an AI assistant that answers questions using only the provided context.\n"
+            "- Do NOT copy large passages from the context.\n"
+            "- Do NOT mention file names or sources in your answer.\n"
+            "- If the answer cannot be found in the context, reply exactly with: "
             "\"I don't know based on the provided documents.\"\n\n"
             f"Context:\n{combined_text}\n\n"
             f"Question: {question}\n\n"
-            "Answer:"
+            "Answer in 1–3 concise sentences:"
         )
 
         try:
@@ -539,7 +543,7 @@ with gr.Blocks(title=CONFIG["client"]["name"]) as demo:
             txt = gr.Textbox(
                 show_label=False,
                 placeholder="Ask a question about your documents and press Enter to send...",
-                lines=2,
+                lines=1,
             )
 
         with gr.Row():