deviandice/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

drop full layers in layerdrop, not half layers

Browse Source
This commit is contained in:
James Betker 2022-03-23 17:15:08 -06:00
parent 57da6d0ddf
commit 8707a3e0c3
2 changed files with 12 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
codes/models/audio/tts/diffusion_encoder.py
View File

@ -152,6 +152,7 @@ class TimestepEmbeddingAttentionLayers(AttentionLayers):
layer_types = default_block * depth layer_types = default_block * depth
self.layer_types = layer_types self.layer_types = layer_types
self.num_layer_types = len(set(self.layer_types))
self.num_attn_layers = len(list(filter(equals('a'), layer_types))) self.num_attn_layers = len(list(filter(equals('a'), layer_types)))
# calculate token shifting # calculate token shifting
@ -225,21 +226,24 @@ class TimestepEmbeddingAttentionLayers(AttentionLayers):
rotary_pos_emb = self.rotary_pos_emb(max_rotary_emb_length, x.device) rotary_pos_emb = self.rotary_pos_emb(max_rotary_emb_length, x.device)
unused_params = [] unused_params = []
to_drop = 0
for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)): for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)):
is_last = ind == (len(self.layers) - 1) if layer_type == 'a':
# Do layer drop where applicable. Do not drop first layer. When doing layer-drop, drop all of the joined layers (e.g. attention + context + feedforward)
if self.training and self.layerdrop_percent > 0 and ind != 0 and random.random() < self.layerdrop_percent:
to_drop = self.num_layer_types
# Do layer drop where applicable. Do not drop first and last layers. hiddens.append(x)
if self.training and self.layerdrop_percent > 0 and not is_last and ind != 0 and random.random() < self.layerdrop_percent: layer_mem = mems.pop(0) if mems else None
if to_drop > 0:
to_drop -= 1
# Record the unused parameters so they can be used in null-operations later to not trigger DDP. # Record the unused parameters so they can be used in null-operations later to not trigger DDP.
unused_params.extend(list(block.parameters())) unused_params.extend(list(block.parameters()))
unused_params.extend(list(residual_fn.parameters())) unused_params.extend(list(residual_fn.parameters()))
unused_params.extend(list(norm.parameters())) unused_params.extend(list(norm.parameters()))
continue continue
if layer_type == 'a':
hiddens.append(x)
layer_mem = mems.pop(0) if mems else None
residual = x residual = x
pre_branch_norm, post_branch_norm, post_main_norm = norm pre_branch_norm, post_branch_norm, post_main_norm = norm

2
codes/models/audio/tts/unet_diffusion_tts_flat.py
View File

@ -21,7 +21,7 @@ class DiffusionTtsFlat(nn.Module):
def __init__( def __init__(
self, self,
model_channels=512, model_channels=512,
num_layers=8, num_layers=16,
in_channels=100, in_channels=100,
in_latent_channels=512, in_latent_channels=512,
in_tokens=8193, in_tokens=8193,