Fix critical bug in PytorchLARS().step: Undefined variable
This commit is contained in:
parent
f6978ae2a2
commit
4a05df34c2
|
@ -181,7 +181,7 @@ class PytorchLARS(Optimizer):
|
|||
state = self.state[p]
|
||||
d_p = p.grad
|
||||
if weight_decay != 0:
|
||||
d_p = d_p.add(param, alpha=weight_decay)
|
||||
d_p = d_p.add(p, alpha=weight_decay)
|
||||
|
||||
if momentum != 0:
|
||||
buf = state.get("momentum_buffer", None)
|
||||
|
|
Loading…
Reference in New Issue
Block a user