slightly increase repeat_penalty to reduce token repetition

app.py

@@ -103,6 +103,7 @@ def caption_frame(frame):
         messages=messages,
         max_tokens=128,
         temperature=0.1,
+        repeat_penalty=1.1,       # discourage exact token repeats
         stop=["<end_of_utterance>"],
     )
     out = (resp["choices"][0].get("message", {}).get("content") or "").strip()