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

src/transformers/models/autoformer/modeling_autoformer.py

@@ -1198,7 +1198,7 @@ def forward(
             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 +1408,7 @@ def forward(
             # 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

@@ -837,7 +837,7 @@ def forward(
             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 +1090,7 @@ def forward(
             # 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_flax_bart.py

@@ -479,7 +479,7 @@ def __call__(
             if output_hidden_states:
                 all_hidden_states = 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([])
             if not deterministic and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -625,7 +625,7 @@ def __call__(
             if output_hidden_states:
                 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([])
             if not deterministic and (dropout_probability < self.layerdrop):
                 layer_outputs = (None, None, None)
             else:

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -1933,7 +1933,7 @@ def forward(
             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 +2276,7 @@ def forward(
             # 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

@@ -579,7 +579,7 @@ def forward(
             # 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

@@ -767,7 +767,7 @@ def forward(
             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 +1019,7 @@ def forward(
             # 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_flax_blenderbot.py

@@ -470,7 +470,7 @@ def __call__(
             if output_hidden_states:
                 all_hidden_states = 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([])
             if not deterministic and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -619,7 +619,7 @@ def __call__(
             if output_hidden_states:
                 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([])
             if not deterministic and (dropout_probability < self.layerdrop):
                 layer_outputs = (None, None, None)
             else:

src/transformers/models/blenderbot_small/modeling_blenderbot_small.py

@@ -765,7 +765,7 @@ def forward(
             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 +1016,7 @@ def forward(
             # 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_flax_blenderbot_small.py

@@ -482,7 +482,7 @@ def __call__(
             if output_hidden_states:
                 all_hidden_states = 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([])
             if not deterministic and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -631,7 +631,7 @@ def __call__(
             if output_hidden_states:
                 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([])
             if not deterministic and (dropout_probability < self.layerdrop):
                 layer_outputs = (None, None, None)
             else:

src/transformers/models/conditional_detr/modeling_conditional_detr.py

@@ -1224,7 +1224,7 @@ def forward(
             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 +1378,7 @@ def forward(
             # 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 @@ def forward(
                 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

@@ -979,7 +979,7 @@ def forward(
             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 +1118,7 @@ def forward(
             # 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

@@ -580,7 +580,7 @@ def forward(
         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

@@ -550,7 +550,7 @@ def forward(
                 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 +794,7 @@ def forward(
                 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 @@ def forward(
                 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 @@ def forward(
                 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

@@ -1205,7 +1205,7 @@ def forward(
             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 +1425,7 @@ def forward(
             # 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

@@ -1871,7 +1871,7 @@ def forward(
             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 +2135,7 @@ def forward(
             # 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

@@ -813,7 +813,7 @@ def forward(
                 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 +1057,7 @@ def forward(
                 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_flax_marian.py

@@ -493,7 +493,7 @@ def __call__(
             if output_hidden_states:
                 all_hidden_states = 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([])
             if not deterministic and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -642,7 +642,7 @@ def __call__(
             if output_hidden_states:
                 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([])
             if not deterministic and (dropout_probability < self.layerdrop):
                 layer_outputs = (None, None, None)
             else:

src/transformers/models/marian/modeling_marian.py

@@ -778,7 +778,7 @@ def forward(
             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 +1024,7 @@ def forward(
             # 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

@@ -1862,7 +1862,7 @@ def forward(
             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

@@ -764,7 +764,7 @@ def forward(
             # 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_flax_mbart.py

@@ -492,7 +492,7 @@ def __call__(
             if output_hidden_states:
                 all_hidden_states = 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([])
             if not deterministic and (dropout_probability < self.layerdrop):  # skip the layer
                 layer_outputs = (None, None)
             else:
@@ -640,7 +640,7 @@ def __call__(
             if output_hidden_states:
                 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([])
             if not deterministic and (dropout_probability < self.layerdrop):
                 layer_outputs = (None, None, None)
             else:

src/transformers/models/mbart/modeling_mbart.py

@@ -819,7 +819,7 @@ def forward(
             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 +1074,7 @@ def forward(
             # 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

@@ -610,7 +610,7 @@ def forward(
                 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: