[`TF`] Also apply patch to support left padding (#25085)

* tf versions

* apply changes to other models

* 3 models slipped through the cracks

src/transformers/models/ctrl/modeling_ctrl.py

@@ -785,7 +785,9 @@ class CTRLForSequenceClassification(CTRLPreTrainedModel):
             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.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/ctrl/modeling_tf_ctrl.py

@@ -798,16 +798,10 @@ class TFCTRLForSequenceClassification(TFCTRLPreTrainedModel, TFSequenceClassific
         else:
             if input_ids is not None:
                 sequence_lengths = (
-                    tf.reduce_sum(
-                        tf.cast(
-                            tf.math.not_equal(input_ids, self.config.pad_token_id),
-                            dtype=input_ids.dtype,
-                        ),
-                        -1,
-                        keepdims=False,
-                    )
+                    tf.argmax(tf.cast(tf.math.equal(input_ids, self.config.pad_token_id), input_ids.dtype), axis=-1)
                     - 1
                 )
+                sequence_lengths = tf.where(sequence_lengths >= 0, sequence_lengths, input_ids.shape[-1] - 1)
                 in_logits = tf.gather(logits, sequence_lengths, batch_dims=1, axis=1)
             else:
                 sequence_lengths = -1

src/transformers/models/gpt2/modeling_tf_gpt2.py

@@ -1082,16 +1082,10 @@ class TFGPT2ForSequenceClassification(TFGPT2PreTrainedModel, TFSequenceClassific
         else:
             if input_ids is not None:
                 sequence_lengths = (
-                    tf.reduce_sum(
-                        tf.cast(
-                            tf.math.not_equal(input_ids, self.config.pad_token_id),
-                            dtype=input_ids.dtype,
-                        ),
-                        -1,
-                        keepdims=False,
-                    )
+                    tf.argmax(tf.cast(tf.math.equal(input_ids, self.config.pad_token_id), input_ids.dtype), axis=-1)
                     - 1
                 )
+                sequence_lengths = tf.where(sequence_lengths >= 0, sequence_lengths, input_ids.shape[-1] - 1)
                 in_logits = tf.gather(logits, sequence_lengths, batch_dims=1, axis=1)
             else:
                 sequence_lengths = -1

src/transformers/models/gptj/modeling_tf_gptj.py

@@ -867,16 +867,10 @@ class TFGPTJForSequenceClassification(TFGPTJPreTrainedModel, TFSequenceClassific
         else:
             if input_ids is not None:
                 sequence_lengths = (
-                    tf.reduce_sum(
-                        tf.cast(
-                            tf.math.not_equal(input_ids, self.config.pad_token_id),
-                            dtype=input_ids.dtype,
-                        ),
-                        -1,
-                        keepdims=False,
-                    )
+                    tf.argmax(tf.cast(tf.math.equal(input_ids, self.config.pad_token_id), input_ids.dtype), axis=-1)
                     - 1
                 )
+                sequence_lengths = tf.where(sequence_lengths >= 0, sequence_lengths, input_ids.shape[-1] - 1)
                 in_logits = tf.gather(logits, sequence_lengths, batch_dims=1, axis=1)
             else:
                 sequence_lengths = -1

src/transformers/models/openai/modeling_openai.py

@@ -813,7 +813,9 @@ class OpenAIGPTForSequenceClassification(OpenAIGPTPreTrainedModel):
             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.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(

src/transformers/models/openai/modeling_tf_openai.py

@@ -809,16 +809,10 @@ class TFOpenAIGPTForSequenceClassification(TFOpenAIGPTPreTrainedModel, TFSequenc
         else:
             if input_ids is not None:
                 sequence_lengths = (
-                    tf.reduce_sum(
-                        tf.cast(
-                            tf.math.not_equal(input_ids, self.config.pad_token_id),
-                            dtype=input_ids.dtype,
-                        ),
-                        -1,
-                        keepdims=False,
-                    )
+                    tf.argmax(tf.cast(tf.math.equal(input_ids, self.config.pad_token_id), input_ids.dtype), axis=-1)
                     - 1
                 )
+                sequence_lengths = tf.where(sequence_lengths >= 0, sequence_lengths, input_ids.shape[-1] - 1)
                 in_logits = tf.gather(logits, sequence_lengths, batch_dims=1, axis=1)
             else:
                 sequence_lengths = -1

src/transformers/models/transfo_xl/modeling_tf_transfo_xl.py

@@ -1066,16 +1066,10 @@ class TFTransfoXLForSequenceClassification(TFTransfoXLPreTrainedModel, TFSequenc
         else:
             if input_ids is not None:
                 sequence_lengths = (
-                    tf.reduce_sum(
-                        tf.cast(
-                            tf.math.not_equal(input_ids, self.config.pad_token_id),
-                            dtype=input_ids.dtype,
-                        ),
-                        -1,
-                        keepdims=False,
-                    )
+                    tf.argmax(tf.cast(tf.math.equal(input_ids, self.config.pad_token_id), input_ids.dtype), axis=-1)
                     - 1
                 )
+                sequence_lengths = tf.where(sequence_lengths >= 0, sequence_lengths, input_ids.shape[-1] - 1)
                 in_logits = tf.gather(logits, sequence_lengths, batch_dims=1, axis=1)
             else:
                 sequence_lengths = -1

src/transformers/models/transfo_xl/modeling_transfo_xl.py

@@ -1247,7 +1247,9 @@ class TransfoXLForSequenceClassification(TransfoXLPreTrainedModel):
             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.eq(input_ids, self.config.pad_token_id).long().argmax(-1) - 1).to(
+                    logits.device
+                )
             else:
                 sequence_lengths = -1
                 logger.warning(