caffe 0.17 l2 norm grows to inf

[JohnnyHan]

ubuntu 16.04.4
cuda v8.0.16
gtx 1080 ti

prototxt
default_forward_type: FLOAT16
default_backward_type: FLOAT16
default_forward_math: FLOAT16
default_backward_marh: FLOAT16
global_grad_scale: 0.09
global_grad_scale_adaptive:true

solver.prototxt
clip_gradients:150

A auto_encoder net, IN BVLC caffe, l2 norm value less than 500, but nvcaffe0.17,the l2 norm grow slowly from 150 to inf, then i get the nan loss.

FLOAT32 format is the same as FLOAT16
if not setting global_grad_scale_adaptive:true and global_grad_scale: 0.09 , l2 norm grows more quickly to inf

[drnikolaev]

@JohnnyHan can you please try

global_grad_scale: 1
global_grad_scale_adaptive: true

also try to remove clip_gradients.
If it still breaks please attach complete log here. Thank you.