Hi friends, I have just found that the inference speed dropped to only 1/10 of the original model. Had anyone encountered this? Thank you.