from langchain.chains.question_answering import load_qa_chain

chain=load_qa_chain(
    llm=llm,
    #reduce_llm=llm,
    chain_type="map_reduce", 
    question_prompt=question_prompt,
    combine_prompt=combine_prompt)


query = "What is .......?"

#Running function without in-built asynchonous method:
res_a = chain.invoke({"input_documents":documents,"question":query}, return_only_outputs=False, verbose=True)

#Running function with in-built asynchonous method:
res_a = await chain.ainvoke({"input_documents":documents,"question":query}, return_only_outputs=True)

There is no difference in inference time when running this on AWS sagemaker notebook with GPU. Is there any reason for this?