llm = HuggingFaceLLM(
    context_window=4096,
    max_new_tokens=256,
    generate_kwargs={"temperature": 0.7, "do_sample": False},
    system_prompt=system_prompt,
    query_wrapper_prompt=query_wrapper_prompt,
    tokenizer_name="facebook/opt-350m",
    model_name="facebook/opt-350m",
    device_map="auto",
    stopping_ids=[50278, 50279, 50277, 1, 0],
    tokenizer_kwargs={"max_length": 4096},
    # uncomment this if using CUDA to reduce memory usage
    # model_kwargs={"torch_dtype": torch.float16}
)
documents = SimpleDirectoryReader("./documents").load_data()
service_context = ServiceContext.from_defaults(chunk_size=1024, llm=llm)
index = VectorStoreIndex.from_documents(
    documents, service_context=service_context)