Fix typing errors in `Qwen2ForTokenClassification` (#31440)

* Update modeling_qwen2.py

* Fix llama

* More fixes

src/transformers/models/llama/modeling_llama.py

@@ -1545,7 +1545,7 @@ class LlamaForTokenClassification(LlamaPreTrainedModel):
     @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1555,7 +1555,7 @@ class LlamaForTokenClassification(LlamaPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/mistral/modeling_mistral.py

@@ -1480,7 +1480,7 @@ class MistralForTokenClassification(MistralPreTrainedModel):
     @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1490,7 +1490,7 @@ class MistralForTokenClassification(MistralPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/mixtral/modeling_mixtral.py

@@ -1628,7 +1628,7 @@ class MixtralForTokenClassification(MixtralPreTrainedModel):
     @add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1638,7 +1638,7 @@ class MixtralForTokenClassification(MixtralPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/persimmon/modeling_persimmon.py

@@ -1051,7 +1051,7 @@ class PersimmonForTokenClassification(PersimmonPreTrainedModel):
     @add_start_docstrings_to_model_forward(PERSIMMON_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1061,7 +1061,7 @@ class PersimmonForTokenClassification(PersimmonPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/qwen2/modeling_qwen2.py

@@ -1418,7 +1418,7 @@ class Qwen2ForTokenClassification(Qwen2PreTrainedModel):
     @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1428,7 +1428,7 @@ class Qwen2ForTokenClassification(Qwen2PreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/qwen2_moe/modeling_qwen2_moe.py

@@ -1614,7 +1614,7 @@ class Qwen2MoeForTokenClassification(Qwen2MoePreTrainedModel):
     @add_start_docstrings_to_model_forward(QWEN2MOE_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1624,7 +1624,7 @@ class Qwen2MoeForTokenClassification(Qwen2MoePreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/stablelm/modeling_stablelm.py

@@ -1427,7 +1427,7 @@ class StableLmForTokenClassification(StableLmPreTrainedModel):
     @add_start_docstrings_to_model_forward(STABLELM_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1437,7 +1437,7 @@ class StableLmForTokenClassification(StableLmPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

src/transformers/models/starcoder2/modeling_starcoder2.py

@@ -1402,7 +1402,7 @@ class Starcoder2ForTokenClassification(Starcoder2PreTrainedModel):
     @add_start_docstrings_to_model_forward(STARCODER2_INPUTS_DOCSTRING)
     def forward(
         self,
-        input_ids: torch.LongTensor = None,
+        input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_values: Optional[List[torch.FloatTensor]] = None,
@@ -1412,7 +1412,7 @@ class Starcoder2ForTokenClassification(Starcoder2PreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+    ) -> Union[Tuple, TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,