refine cpu (#5214)

* refine transformer cpu

* delete run_pretrain.sh

PaddleNLP/benchmark/transformer/dygraph/train.py

@@ -41,6 +41,7 @@ def do_train(args):
     else:
         rank = 0
         trainer_count = 1
+        paddle.set_device("cpu")
 
     if trainer_count > 1:
         dist.init_parallel_env()

PaddleNLP/benchmark/transformer/static/run_pretrain.sh

-
-python -m paddle.distributed.launch \
-     --gpus="0,1" \
-     train.py

PaddleNLP/benchmark/transformer/static/train.py

@@ -44,8 +44,11 @@ def do_train(args):
             gpu_id) if args.use_gpu else paddle.static.cpu_places()
         trainer_count = 1 if args.use_gpu else len(places)
     else:
-        places = paddle.static.cuda_places(
-        ) if args.use_gpu else paddle.static.cpu_places()
+        if args.use_gpu:
+            places = paddle.static.cuda_places()
+        else:
+            places = paddle.static.cpu_places()
+            paddle.set_device("cpu")
         trainer_count = len(places)
 
     # Set seed for CE

PaddleNLP/examples/machine_translation/transformer/train.py

@@ -33,6 +33,7 @@ def do_train(args):
     else:
         rank = 0
         trainer_count = 1
+        paddle.set_device("cpu")
 
     if trainer_count > 1:
         dist.init_parallel_env()