How to train yoloV3 in FP16 or uint8?

I tried to train yolov3 in FP16.This is the modification process(https://blog.csdn.net/bingpoyinhui/article/details/102782260),but there was an error reported by many users.
the error info:
mxnet.base.MXNetError: Error in operator yolov3loss0_broadcast_mul1: [10:11:00] src/operator/contrib/…/elemwise_op_common.h:135: Check failed: assign(&dattr, vec.at(i)): Incompatible attr in node yolov3loss0_broadcast_mul1 at 1-th input: expected float16, got float32

This error has been reported by many people. Is there any specific solution?thanks