You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have some problem with training with my codes without strategy=ddp_find_unused_parameters_true. When I turned on the flag, it seems there is no parameters that didn't participate in the training. But when I turned off, it always crashed with some error logs like your model has unused parameters ~~~.
You can see attached codes, and I found that the actual error is caused by .detach() operation, because when I removed that there is no problem with training. How can I solve this problem?
Bug description
I have some problem with training with my codes without
strategy=ddp_find_unused_parameters_true
. When I turned on the flag, it seems there is no parameters that didn't participate in the training. But when I turned off, it always crashed with some error logs like your model has unused parameters ~~~.You can see attached codes, and I found that the actual error is caused by
.detach()
operation, because when I removed that there is no problem with training. How can I solve this problem?What version are you seeing the problem on?
v2.2
How to reproduce the bug
Error messages and logs
Environment
Current environment
More info
No response
The text was updated successfully, but these errors were encountered: