Datasets:

ArXiv:
diffusers-benchmarking-bot commited on
Commit
5684612
·
verified ·
1 Parent(s): 70dd1e4

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. main/matryoshka.py +5 -74
main/matryoshka.py CHANGED
@@ -80,7 +80,6 @@ from diffusers.utils import (
80
  USE_PEFT_BACKEND,
81
  BaseOutput,
82
  deprecate,
83
- is_torch_version,
84
  is_torch_xla_available,
85
  logging,
86
  replace_example_docstring,
@@ -869,23 +868,7 @@ class CrossAttnDownBlock2D(nn.Module):
869
 
870
  for i, (resnet, attn) in enumerate(blocks):
871
  if torch.is_grad_enabled() and self.gradient_checkpointing:
872
-
873
- def create_custom_forward(module, return_dict=None):
874
- def custom_forward(*inputs):
875
- if return_dict is not None:
876
- return module(*inputs, return_dict=return_dict)
877
- else:
878
- return module(*inputs)
879
-
880
- return custom_forward
881
-
882
- ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
883
- hidden_states = torch.utils.checkpoint.checkpoint(
884
- create_custom_forward(resnet),
885
- hidden_states,
886
- temb,
887
- **ckpt_kwargs,
888
- )
889
  hidden_states = attn(
890
  hidden_states,
891
  encoder_hidden_states=encoder_hidden_states,
@@ -1030,17 +1013,6 @@ class UNetMidBlock2DCrossAttn(nn.Module):
1030
  hidden_states = self.resnets[0](hidden_states, temb)
1031
  for attn, resnet in zip(self.attentions, self.resnets[1:]):
1032
  if torch.is_grad_enabled() and self.gradient_checkpointing:
1033
-
1034
- def create_custom_forward(module, return_dict=None):
1035
- def custom_forward(*inputs):
1036
- if return_dict is not None:
1037
- return module(*inputs, return_dict=return_dict)
1038
- else:
1039
- return module(*inputs)
1040
-
1041
- return custom_forward
1042
-
1043
- ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
1044
  hidden_states = attn(
1045
  hidden_states,
1046
  encoder_hidden_states=encoder_hidden_states,
@@ -1049,12 +1021,7 @@ class UNetMidBlock2DCrossAttn(nn.Module):
1049
  encoder_attention_mask=encoder_attention_mask,
1050
  return_dict=False,
1051
  )[0]
1052
- hidden_states = torch.utils.checkpoint.checkpoint(
1053
- create_custom_forward(resnet),
1054
- hidden_states,
1055
- temb,
1056
- **ckpt_kwargs,
1057
- )
1058
  else:
1059
  hidden_states = attn(
1060
  hidden_states,
@@ -1192,23 +1159,7 @@ class CrossAttnUpBlock2D(nn.Module):
1192
  hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
1193
 
1194
  if torch.is_grad_enabled() and self.gradient_checkpointing:
1195
-
1196
- def create_custom_forward(module, return_dict=None):
1197
- def custom_forward(*inputs):
1198
- if return_dict is not None:
1199
- return module(*inputs, return_dict=return_dict)
1200
- else:
1201
- return module(*inputs)
1202
-
1203
- return custom_forward
1204
-
1205
- ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
1206
- hidden_states = torch.utils.checkpoint.checkpoint(
1207
- create_custom_forward(resnet),
1208
- hidden_states,
1209
- temb,
1210
- **ckpt_kwargs,
1211
- )
1212
  hidden_states = attn(
1213
  hidden_states,
1214
  encoder_hidden_states=encoder_hidden_states,
@@ -1282,10 +1233,6 @@ class MatryoshkaTransformer2DModel(LegacyModelMixin, LegacyConfigMixin):
1282
  ]
1283
  )
1284
 
1285
- def _set_gradient_checkpointing(self, module, value=False):
1286
- if hasattr(module, "gradient_checkpointing"):
1287
- module.gradient_checkpointing = value
1288
-
1289
  def forward(
1290
  self,
1291
  hidden_states: torch.Tensor,
@@ -1365,19 +1312,8 @@ class MatryoshkaTransformer2DModel(LegacyModelMixin, LegacyConfigMixin):
1365
  # Blocks
1366
  for block in self.transformer_blocks:
1367
  if torch.is_grad_enabled() and self.gradient_checkpointing:
1368
-
1369
- def create_custom_forward(module, return_dict=None):
1370
- def custom_forward(*inputs):
1371
- if return_dict is not None:
1372
- return module(*inputs, return_dict=return_dict)
1373
- else:
1374
- return module(*inputs)
1375
-
1376
- return custom_forward
1377
-
1378
- ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
1379
- hidden_states = torch.utils.checkpoint.checkpoint(
1380
- create_custom_forward(block),
1381
  hidden_states,
1382
  attention_mask,
1383
  encoder_hidden_states,
@@ -1385,7 +1321,6 @@ class MatryoshkaTransformer2DModel(LegacyModelMixin, LegacyConfigMixin):
1385
  timestep,
1386
  cross_attention_kwargs,
1387
  class_labels,
1388
- **ckpt_kwargs,
1389
  )
1390
  else:
1391
  hidden_states = block(
@@ -2724,10 +2659,6 @@ class MatryoshkaUNet2DConditionModel(
2724
  for module in self.children():
2725
  fn_recursive_set_attention_slice(module, reversed_slice_size)
2726
 
2727
- def _set_gradient_checkpointing(self, module, value=False):
2728
- if hasattr(module, "gradient_checkpointing"):
2729
- module.gradient_checkpointing = value
2730
-
2731
  def enable_freeu(self, s1: float, s2: float, b1: float, b2: float):
2732
  r"""Enables the FreeU mechanism from https://arxiv.org/abs/2309.11497.
2733
 
 
80
  USE_PEFT_BACKEND,
81
  BaseOutput,
82
  deprecate,
 
83
  is_torch_xla_available,
84
  logging,
85
  replace_example_docstring,
 
868
 
869
  for i, (resnet, attn) in enumerate(blocks):
870
  if torch.is_grad_enabled() and self.gradient_checkpointing:
871
+ hidden_states = self._gradient_checkpointing_func(resnet, hidden_states, temb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
872
  hidden_states = attn(
873
  hidden_states,
874
  encoder_hidden_states=encoder_hidden_states,
 
1013
  hidden_states = self.resnets[0](hidden_states, temb)
1014
  for attn, resnet in zip(self.attentions, self.resnets[1:]):
1015
  if torch.is_grad_enabled() and self.gradient_checkpointing:
 
 
 
 
 
 
 
 
 
 
 
1016
  hidden_states = attn(
1017
  hidden_states,
1018
  encoder_hidden_states=encoder_hidden_states,
 
1021
  encoder_attention_mask=encoder_attention_mask,
1022
  return_dict=False,
1023
  )[0]
1024
+ hidden_states = self._gradient_checkpointing_func(resnet, hidden_states, temb)
 
 
 
 
 
1025
  else:
1026
  hidden_states = attn(
1027
  hidden_states,
 
1159
  hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
1160
 
1161
  if torch.is_grad_enabled() and self.gradient_checkpointing:
1162
+ hidden_states = self._gradient_checkpointing_func(resnet, hidden_states, temb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1163
  hidden_states = attn(
1164
  hidden_states,
1165
  encoder_hidden_states=encoder_hidden_states,
 
1233
  ]
1234
  )
1235
 
 
 
 
 
1236
  def forward(
1237
  self,
1238
  hidden_states: torch.Tensor,
 
1312
  # Blocks
1313
  for block in self.transformer_blocks:
1314
  if torch.is_grad_enabled() and self.gradient_checkpointing:
1315
+ hidden_states = self._gradient_checkpointing_func(
1316
+ block,
 
 
 
 
 
 
 
 
 
 
 
1317
  hidden_states,
1318
  attention_mask,
1319
  encoder_hidden_states,
 
1321
  timestep,
1322
  cross_attention_kwargs,
1323
  class_labels,
 
1324
  )
1325
  else:
1326
  hidden_states = block(
 
2659
  for module in self.children():
2660
  fn_recursive_set_attention_slice(module, reversed_slice_size)
2661
 
 
 
 
 
2662
  def enable_freeu(self, s1: float, s2: float, b1: float, b2: float):
2663
  r"""Enables the FreeU mechanism from https://arxiv.org/abs/2309.11497.
2664