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

Reduce min of scale in Conv2dSubsampling from 0.01 to 0.2

Browse Source
This commit is contained in:
Daniel Povey 2022-11-29 16:18:41 +08:00
parent 441fcf063d
commit 7589e37689
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
View File

@ -1720,7 +1720,7 @@ class Conv2dSubsampling(nn.Module):
self.scale = nn.Parameter(torch.ones(out_height * layer3_channels))
self.scale_max = 1.0
self.scale_min = ScheduledFloat((0.0, 0.9), (4000.0, 0.01))
self.scale_min = ScheduledFloat((0.0, 0.9), (4000.0, 0.2))
self.out = LinearWithAuxLoss(out_height * layer3_channels, out_channels,
aux_grad_scale=_aux_grad_scale(), prob=_aux_grad_prob_out())