Remove out_balancer of NonlinAttentionModule

This commit is contained in:
Daniel Povey 2022-11-23 18:37:45 +08:00
parent 9138695dfe
commit b88f12fe83

View File

@ -1443,16 +1443,6 @@ class NonlinAttentionModule(nn.Module):
initial_scale=0.05)
# put quite strict limits on the min_positive and max_positive at the output,
# because we noticed that poorly-trained instances of NonlinAttentionModule seem
# to have a larger mean-offset at the output for some reason.
self.out_balancer = ActivationBalancer(
channels, channel_dim=-1,
min_positive=0.45, max_positive=0.55,
min_abs=0.005, max_abs=1.0,
min_prob=0.05,
)
def forward(self,
x: Tensor,