Update app.py

app.py

@@ -44,12 +44,14 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
 
     thinking_prompt = messages + [{
         "role": "user",
-        "content": f"{message}\n\nThinking a short step-by-step before answering."
+        "content": f"{message}\n\nThink step-by-step before answering."
     }]
 
     reasoning = ""
     yield '<div class="markdown-think">Thinking...</div>'
 
+    start = time.time()
+
     for chunk in client.chat_completion(
         thinking_prompt,
         max_tokens=max_tokens,
@@ -62,7 +64,14 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
         styled_thought = f'<div class="markdown-think">{reasoning.strip()}</div>'
         yield styled_thought
 
-    time.sleep(0.5)
+    elapsed = time.time() - start
+
+    yield f"""
+    <div style="margin-top:12px;padding:8px 12px;background-color:#222;border-left:4px solid #888;
+                font-family:'JetBrains Mono', monospace;color:#ccc;font-size:14px;">
+        Pensou por {elapsed:.1f} segundos
+    </div>
+    """
 
     final_prompt = messages + [
         {"role": "user", "content": message},