How to fix certain parameters during gluon training?

Gluon
1

Setattr(“grad_req”, “null”) works only for setting gradient of a block of parameters. I want to set “grad_req”, “null” for a particular parameter of the network.

2

I think you can reference to this post

2 Likes
3

Is there a way to fix running_mean and running_var and not compute it based on the current batch?

4

Maybe you can set use_global_stats=True ?

2 Likes