mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-12-11 06:55:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'scaled_adam_exp760' into scaled_adam_exp765

Browse Source
This commit is contained in:
Daniel Povey 2022-12-23 13:08:09 +08:00
commit cff350d8de
1 changed files with 13 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
View File

@ -1712,10 +1712,13 @@ class ConvNeXt(nn.Module):
self.out_balancer = ActivationBalancer(
channels, channel_dim=1,
min_positive=0.5, max_positive=0.5,
min_positive=0.4, max_positive=0.6,
min_abs=0.25, max_abs=6.0,
)
self.out_whiten = Whiten(num_groups=1,
whitening_limit=5.0,
prob=(0.025, 0.25),
grad_scale=0.01)
def forward(self, x: Tensor) -> Tensor:
@ -1739,6 +1742,10 @@ class ConvNeXt(nn.Module):
x = bypass + x
x = self.out_balancer(x)
x = x.transpose(1, 3) # (N, W, H, C); need channel dim to be last
x = self.out_whiten(x)
x = x.transpose(1, 3) # (N, C, H, W)
return x
@ -1845,13 +1852,13 @@ class Conv2dSubsampling(nn.Module):
self.scale_min = ScheduledFloat((0.0, 0.9), (4000.0, 0.1))
self.out = nn.Linear(out_width * layer3_channels, out_channels)
# use a much larger than normal grad_scale on this whitening module;
# there is only one such module, so there is not a concern about adding
# together many copies of this extra gradient term.
# use a larger than normal grad_scale on this whitening module; there is
# only one such module, so there is not a concern about adding together
# many copies of this extra gradient term.
self.out_whiten = Whiten(num_groups=1,
whitening_limit=_whitening_schedule(4.0),
prob=(0.025, 0.25),
grad_scale=0.05)
grad_scale=0.02)
self.out_norm = BasicNorm(out_channels)
self.dropout = Dropout2(dropout)