[ `ForSequenceClassification`] Support `left` padding (#24979)

* support left padding

* nit

* Update src/transformers/models/gpt_neox/modeling_gpt_neox.py

* Update src/transformers/models/gpt_neox/modeling_gpt_neox.py

src/transformers/models/deprecated/open_llama/modeling_open_llama.py

@@ -956,7 +956,9 @@ class OpenLlamaForSequenceClassification(OpenLlamaPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
 

src/transformers/models/gpt2/modeling_gpt2.py

@@ -1443,7 +1443,9 @@ class GPT2ForSequenceClassification(GPT2PreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py

@@ -934,7 +934,9 @@ class GPTBigCodeForSequenceClassification(GPTBigCodePreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -878,7 +878,9 @@ class GPTNeoForSequenceClassification(GPTNeoPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gpt_neox/modeling_gpt_neox.py

@@ -926,7 +926,9 @@ class GPTNeoXForSequenceClassification(GPTNeoXPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/gptj/modeling_gptj.py

@@ -1002,7 +1002,9 @@ class GPTJForSequenceClassification(GPTJPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/llama/modeling_llama.py

@@ -971,7 +971,9 @@ class LlamaForSequenceClassification(LlamaPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
 

src/transformers/models/opt/modeling_opt.py

@@ -1084,7 +1084,9 @@ class OPTForSequenceClassification(OPTPreTrainedModel):
             sequence_lengths = -1
         else:
             if input_ids is not None:
-                sequence_lengths = (torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1).to(logits.device)
+                sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(