Replace python random with torch.rand to enable dynamo.export (#24434)

* Replace python random with torch.rand to enable dynamo.export

* revert changes to flax model code

* Remove unused random import

* Fix torch template

* Move torch.manual_seed(0) to right location

src/transformers/models/autoformer/modeling_autoformer.py

@@ -17,7 +17,6 @@
 """ PyTorch Autoformer model."""
 
 import math
-import random
 from dataclasses import dataclass
 from typing import List, Optional, Tuple, Union
 
@@ -1198,7 +1197,7 @@ class AutoformerEncoder(AutoformerPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1408,7 +1407,7 @@ class AutoformerDecoder(AutoformerPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/bart/modeling_bart.py

@@ -15,7 +15,6 @@
 """ PyTorch BART model."""
 import copy
 import math
-import random
 import warnings
 from typing import List, Optional, Tuple, Union
 
@@ -837,7 +836,7 @@ class BartEncoder(BartPretrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1090,7 +1089,7 @@ class BartDecoder(BartPretrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -17,7 +17,6 @@
 
 import copy
 import math
-import random
 from typing import List, Optional, Tuple, Union
 
 import numpy as np
@@ -1933,7 +1932,7 @@ class BigBirdPegasusEncoder(BigBirdPegasusPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -2276,7 +2275,7 @@ class BigBirdPegasusDecoder(BigBirdPegasusPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/biogpt/modeling_biogpt.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 from typing import Optional, Tuple, Union
 
 import torch
@@ -579,7 +578,7 @@ class BioGptModel(BioGptPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/blenderbot/modeling_blenderbot.py

@@ -18,7 +18,6 @@
 import copy
 import math
 import os
-import random
 import warnings
 from typing import List, Optional, Tuple, Union
 
@@ -767,7 +766,7 @@ class BlenderbotEncoder(BlenderbotPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1019,7 +1018,7 @@ class BlenderbotDecoder(BlenderbotPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/blenderbot_small/modeling_blenderbot_small.py

@@ -17,7 +17,6 @@
 
 import copy
 import math
-import random
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -765,7 +764,7 @@ class BlenderbotSmallEncoder(BlenderbotSmallPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1016,7 +1015,7 @@ class BlenderbotSmallDecoder(BlenderbotSmallPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/conditional_detr/modeling_conditional_detr.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 from dataclasses import dataclass
 from typing import Dict, List, Optional, Tuple
 
@@ -1224,7 +1223,7 @@ class ConditionalDetrEncoder(ConditionalDetrPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1378,7 +1377,7 @@ class ConditionalDetrDecoder(ConditionalDetrPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
             if idx == 0:

src/transformers/models/data2vec/modeling_data2vec_audio.py

@@ -587,7 +587,7 @@ class Data2VecAudioEncoder(nn.Module):
                 all_hidden_states = all_hidden_states + (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = np.random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.config.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled:

src/transformers/models/detr/modeling_detr.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 from dataclasses import dataclass
 from typing import Dict, List, Optional, Tuple
 
@@ -979,7 +978,7 @@ class DetrEncoder(DetrPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1118,7 +1117,7 @@ class DetrDecoder(DetrPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/flaubert/modeling_flaubert.py

@@ -16,7 +16,6 @@
 
 import itertools
 import math
-import random
 from dataclasses import dataclass
 from typing import Dict, Optional, Tuple, Union
 
@@ -580,7 +579,7 @@ class FlaubertModel(FlaubertPreTrainedModel):
         attentions = () if output_attentions else None
         for i in range(self.n_layers):
             # LayerDrop
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/fsmt/modeling_fsmt.py

@@ -28,7 +28,6 @@
 """PyTorch Fairseq model, ported from https://github.com/pytorch/fairseq/tree/master/examples/wmt19"""
 
 import math
-import random
 from typing import Any, Dict, List, Optional, Tuple, Union
 
 import torch
@@ -550,7 +549,7 @@ class FSMTEncoder(nn.Module):
                 encoder_states += (x,)
                 x = x.transpose(0, 1)  # B x T x C -> T x B x C
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 attn = None
             else:
@@ -794,7 +793,7 @@ class FSMTDecoder(nn.Module):
                 x = x.transpose(0, 1)
                 all_hidden_states += (x,)
                 x = x.transpose(0, 1)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/hubert/modeling_hubert.py

@@ -725,7 +725,7 @@ class HubertEncoder(nn.Module):
                 all_hidden_states = all_hidden_states + (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = np.random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.config.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled:
@@ -814,7 +814,7 @@ class HubertEncoderStableLayerNorm(nn.Module):
                 all_hidden_states = all_hidden_states + (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = np.random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.config.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled:

src/transformers/models/informer/modeling_informer.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """ PyTorch Informer model."""
 
-import random
 from typing import List, Optional, Tuple, Union
 
 import numpy as np
@@ -1205,7 +1204,7 @@ class InformerEncoder(InformerPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1425,7 +1424,7 @@ class InformerDecoder(InformerPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/led/modeling_led.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 import warnings
 from dataclasses import dataclass
 from typing import List, Optional, Tuple, Union
@@ -1871,7 +1870,7 @@ class LEDEncoder(LEDPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None, None)
@@ -2135,7 +2134,7 @@ class LEDDecoder(LEDPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/m2m_100/modeling_m2m_100.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -813,7 +812,7 @@ class M2M100Encoder(M2M100PreTrainedModel):
                 encoder_states = encoder_states + (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled:
@@ -1057,7 +1056,7 @@ class M2M100Decoder(M2M100PreTrainedModel):
                 all_hidden_states += (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled:

src/transformers/models/marian/modeling_marian.py

@@ -17,7 +17,6 @@
 
 import copy
 import math
-import random
 from typing import Dict, List, Optional, Tuple, Union
 
 import numpy as np
@@ -778,7 +777,7 @@ class MarianEncoder(MarianPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1024,7 +1023,7 @@ class MarianDecoder(MarianPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/mask2former/modeling_mask2former.py

@@ -15,7 +15,6 @@
 """ PyTorch Mask2Former model."""
 
 import math
-import random
 import warnings
 from dataclasses import dataclass
 from typing import Dict, List, Optional, Tuple
@@ -1862,7 +1861,7 @@ class Mask2FormerMaskedAttentionDecoder(nn.Module):
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
 
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             if self.training and (dropout_probability < self.layerdrop):
                 continue

src/transformers/models/maskformer/modeling_maskformer.py

@@ -15,7 +15,6 @@
 """ PyTorch MaskFormer model."""
 
 import math
-import random
 from dataclasses import dataclass
 from numbers import Number
 from typing import Dict, List, Optional, Tuple
@@ -764,7 +763,7 @@ class DetrDecoder(nn.Module):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/mbart/modeling_mbart.py

@@ -15,7 +15,6 @@
 """ PyTorch MBART model."""
 import copy
 import math
-import random
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -819,7 +818,7 @@ class MBartEncoder(MBartPreTrainedModel):
             if output_hidden_states:
                 encoder_states = encoder_states + (hidden_states,)
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -1074,7 +1073,7 @@ class MBartDecoder(MBartPreTrainedModel):
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
             if output_hidden_states:
                 all_hidden_states += (hidden_states,)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
             if self.training and (dropout_probability < self.layerdrop):
                 continue
 

src/transformers/models/mctct/modeling_mctct.py

@@ -16,7 +16,6 @@
 
 
 import math
-import random
 from typing import Optional, Tuple, Union
 
 import torch
@@ -610,7 +609,7 @@ class MCTCTEncoder(MCTCTPreTrainedModel):
                 encoder_states = encoder_states + (hidden_states,)
 
             # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
-            dropout_probability = random.uniform(0, 1)
+            dropout_probability = torch.rand([])
 
             skip_the_layer = True if self.training and (dropout_probability < self.config.layerdrop) else False
             if not skip_the_layer or deepspeed_zero3_is_enabled: