Which layers of TensorRT will work in fp16 mode when enable the --half2 option?

I want to modify the caffe to train the fp16 weights for TensorRT, so could anyone tell me which layers of TensorRT will work in fp16 mode when enable the --half2 option? Thanks alot!

Hi,

Please check the duplicate post in tx2 forum.
https://devtalk.nvidia.com/default/topic/999527/jetson-tx2/which-layers-of-tensorrt-will-work-in-fp16-mode-when-enable-the-half2-option-/