Increase layer_skip_rate slightly

This commit is contained in:
Daniel Povey 2023-01-08 15:48:23 +08:00
parent 62b42887b4
commit 326cb75033

View File

@ -180,7 +180,7 @@ class Zipformer(EncoderInterface):
# the model from relying too much on them at the end. (They tend to train
# slowly, so don't increase their skip-rate at the beginning.)
layer_skip_rate = (ScheduledFloat((0.0, 0.5), (4000.0, 0.0), default=0,) +
(0.025 * (downsampling_factor[i] ** 0.5)))
(0.03 * (downsampling_factor[i] ** 0.5)))
encoder_layer = ZipformerEncoderLayer(
embed_dim=encoder_dim[i],
@ -217,7 +217,7 @@ class Zipformer(EncoderInterface):
)
# we are adding a new attribute here.
# this will be interpreted by get_named_parameter_groups_with_lrs().
encoder.lr_scale = downsampling_factor[i] ** -0.25
encoder.lr_scale = downsampling_factor[i] ** -0.33
encoders.append(encoder)
self.encoders = nn.ModuleList(encoders)