mirrors/icefall
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Change default prob on limit_param_value from 0.2 to 0.6.

Browse Source
This commit is contained in:
Daniel Povey 2022-11-27 14:00:59 +08:00
parent 2f4df1278d
commit ff361a7495
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/scaling.py
View File

@ -893,7 +893,7 @@ class LimitParamValue(torch.autograd.Function):
def limit_param_value(x: Tensor, def limit_param_value(x: Tensor,
min: float, max: float, min: float, max: float,
prob: float = 0.2): prob: float = 0.6):
# You apply this to (typically) an nn.Parameter during training to ensure that its # You apply this to (typically) an nn.Parameter during training to ensure that its
# (elements mostly) stays within a supplied range. This is done by modifying the # (elements mostly) stays within a supplied range. This is done by modifying the
# gradients in backprop. # gradients in backprop.