Created by: liyuanhangsc

can we call paddle.infer() in a multi-thread environment and get the right predict value?