reduce ctx and max tokens for performance

app.py

@@ -75,7 +75,7 @@ def load_llm():
     llm = Llama(
         model_path=MODEL_FILE,
         chat_handler=handler,
-        n_ctx=8192,
+        n_ctx=1024,
         verbose=False,
     )
     logging.info("Llama model loaded successfully.")
@@ -129,7 +129,7 @@ def caption_frame(frame):
         logging.debug("Sending chat completion request...")
         resp = llm.create_chat_completion(
             messages=messages,
-            max_tokens=256,
+            max_tokens=128,
             temperature=0.1,
             stop=["<end_of_utterance>"],
         )