Why build fp16 network all input dtype are set trt.float32 not trt.float16?

NVIDIA / trt-samples-for-hackathon-cn

Simple samples for TensorRT programming

Apache License 2.0

1.5k stars 338 forks source link

Closed jackzhou121 closed 1 year ago

jackzhou121 commented 1 year ago

When I try set input dtype to trt.float16, the result is not right. I want to know why?