Update app.py

app.py

@@ -82,7 +82,7 @@ client = InferenceClient(LLM_NAME)
 Settings.llm = HuggingFaceInferenceAPI(model_name=LLM_NAME,
 #                              device_map="auto",
 #                              system_prompt = system_prompt,
-                              context_window=8192,
+                              context_window=6144,
                               max_new_tokens=3072,
 #                              stopping_ids=[50278, 50279, 50277, 1, 0],
                               generate_kwargs={"temperature": temperature, "top_p":top_p, "repetition_penalty": repetition_penalty,