diff --git a/cloud/pcloud_train.sh b/cloud/pcloud_train.sh
index e42da1d629ee5876dbcfe81a1359f30e2c983d1c..75949574d89a4210db5a4dd89dd71f2083a05ae4 100644
--- a/cloud/pcloud_train.sh
+++ b/cloud/pcloud_train.sh
@@ -13,7 +13,7 @@ python ./cloud/split_data.py \
 --in_manifest_path=${DEV_MANIFEST} \
 --out_manifest_path='/local.manifest.dev'
 
-python train.py \
+python -u train.py \
 --batch_size=$BATCH_SIZE \
 --use_gpu=1 \
 --trainer_count=${NUM_GPU} \
@@ -21,4 +21,4 @@ python train.py \
 --is_local=${IS_LOCAL} \
 --train_manifest_path='/local.manifest.train' \
 --dev_manifest_path='/local.manifest.dev' \
---output_model_dir=${MODEL_PATH} \
+--output_model_dir=${MODEL_PATH} 2>&1 | tee ./log/train.log
diff --git a/train.py b/train.py
index 8e95d7bc888316d0433fd331f4fb5c916a6195d3..1d0b92fff7726c4a5fc56757ca53c7619b92ded1 100644
--- a/train.py
+++ b/train.py
@@ -37,12 +37,12 @@ parser.add_argument(
     help="RNN layer number. (default: %(default)s)")
 parser.add_argument(
     "--rnn_layer_size",
-    default=1280,
+    default=1024,
     type=int,
     help="RNN layer cell number. (default: %(default)s)")
 parser.add_argument(
     "--use_gru",
-    default=True,
+    default=False,
     type=bool,
     help="Use GRU or simple RNN. (default: %(default)s)")
 parser.add_argument(