[s2s trainer] fix DP mode (#8823)

* fix DP case on multi-gpu

* make executable

* test all 3 modes

* use the correct check for distributed

* dp doesn't need a special case

* restore original name

* cleanup

examples/seq2seq/seq2seq_trainer.py

@@ -122,7 +122,8 @@ class Seq2SeqTrainer(Trainer):
         else:
             if self.args.sortish_sampler:
                 self.train_dataset.make_sortish_sampler(
-                    self.args.per_device_train_batch_size, distributed=self.args.n_gpu > 1
+                    self.args.per_device_train_batch_size,
+                    distributed=(self.args.local_rank != -1),
                 )
 
             return (