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

Use 2 groups in whitening for NonlinAttentionModule; limit 40->20.

Browse Source
This commit is contained in:
Daniel Povey 2022-11-21 23:23:41 +08:00
parent b3b5e8b9b9
commit 71f118e725
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
View File

@ -1424,8 +1424,8 @@ class NonlinAttentionModule(nn.Module):
)
# give it a high limit, because it is quite high-dimensional and is
# a projection of a lower-dimensional embedding.
self.whiten = Whiten(num_groups=1,
whitening_limit=40.0,
self.whiten = Whiten(num_groups=2,
whitening_limit=20.0,
prob=(0.025, 0.25),
grad_scale=0.01)