Fix CIs for PyTorch 1.13 (huggingface#20686)

* fix 1

* fix 2

* fix 3

* fix 4

Co-authored-by: ydshieh <[email protected]>

src/transformers/models/bart/modeling_bart.py

@@ -1538,7 +1538,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -2738,7 +2738,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

src/transformers/models/bloom/modeling_bloom.py

@@ -1057,7 +1057,7 @@ def forward(
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(dim=-1) - 1
+                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/clip/modeling_clip.py

@@ -734,7 +734,8 @@ def forward(
         # take features from the eot embedding (eot_token is the highest number in each sequence)
         # casting to torch.int for onnx compatibility: argmax doesn't support int64 inputs with opset 14
         pooled_output = last_hidden_state[
-            torch.arange(last_hidden_state.shape[0], device=input_ids.device), input_ids.to(torch.int).argmax(dim=-1)
+            torch.arange(last_hidden_state.shape[0], device=last_hidden_state.device),
+            input_ids.to(dtype=torch.int, device=last_hidden_state.device).argmax(dim=-1),
         ]
 
         if not return_dict:

src/transformers/models/clipseg/modeling_clipseg.py

@@ -746,7 +746,8 @@ def forward(
         # take features from the eot embedding (eot_token is the highest number in each sequence)
         # casting to torch.int for onnx compatibility: argmax doesn't support int64 inputs with opset 14
         pooled_output = last_hidden_state[
-            torch.arange(last_hidden_state.shape[0], device=input_ids.device), input_ids.to(torch.int).argmax(dim=-1)
+            torch.arange(last_hidden_state.shape[0], device=last_hidden_state.device),
+            input_ids.to(dtype=torch.int, device=last_hidden_state.device).argmax(dim=-1),
         ]
 
         if not return_dict:

src/transformers/models/gpt2/modeling_gpt2.py

@@ -1401,7 +1401,7 @@ def forward(
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
+                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -883,7 +883,7 @@ def forward(
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
+                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gptj/modeling_gptj.py

@@ -969,7 +969,7 @@ def forward(
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
+                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/groupvit/modeling_groupvit.py

@@ -1134,7 +1134,8 @@ def forward(
         # take features from the eot embedding (eot_token is the highest number in each sequence)
         # casting to torch.int for onnx compatibility: argmax doesn't support int64 inputs with opset 14
         pooled_output = last_hidden_state[
-            torch.arange(last_hidden_state.shape[0], device=input_ids.device), input_ids.to(torch.int).argmax(dim=-1)
+            torch.arange(last_hidden_state.shape[0], device=last_hidden_state.device),
+            input_ids.to(dtype=torch.int, device=last_hidden_state.device).argmax(dim=-1),
         ]
 
         if not return_dict:

src/transformers/models/led/modeling_led.py

@@ -2608,7 +2608,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

src/transformers/models/mbart/modeling_mbart.py

@@ -1525,7 +1525,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

src/transformers/models/mvp/modeling_mvp.py

@@ -1674,7 +1674,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

src/transformers/models/opt/modeling_opt.py

@@ -1069,7 +1069,7 @@ def forward(
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
+                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/plbart/modeling_plbart.py

@@ -1496,7 +1496,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/modeling_{{cookiecutter.lowercase_modelname}}.py

@@ -2982,7 +2982,7 @@ def forward(
         )
         hidden_states = outputs[0]  # last hidden state
 
-        eos_mask = input_ids.eq(self.config.eos_token_id)
+        eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
 
         if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
             raise ValueError("All examples must have the same number of <eos> tokens.")