diff --git a/deepspeech/exps/u2/bin/test.py b/deepspeech/exps/u2/bin/test.py
index a80106f74..4ab1c647e 100644
--- a/deepspeech/exps/u2/bin/test.py
+++ b/deepspeech/exps/u2/bin/test.py
@@ -53,4 +53,4 @@ if __name__ == "__main__":
     # Setting for profiling
     pr = cProfile.Profile()
     pr.runcall(main, config, args)
-    pr.dump_stats(os.path.join('.', 'test.profile'))
+    pr.dump_stats(os.path.join(args.output, 'train.profile'))
diff --git a/deepspeech/exps/u2/bin/train.py b/deepspeech/exps/u2/bin/train.py
index f23716781..9dd0041dd 100644
--- a/deepspeech/exps/u2/bin/train.py
+++ b/deepspeech/exps/u2/bin/train.py
@@ -56,4 +56,4 @@ if __name__ == "__main__":
     # Setting for profiling
     pr = cProfile.Profile()
     pr.runcall(main, config, args)
-    pr.dump_stats(os.path.join('.', 'train.profile'))
+    pr.dump_stats(os.path.join(args.output, 'train.profile'))
diff --git a/deepspeech/utils/layer_tools.py b/deepspeech/utils/layer_tools.py
index 429391874..67f3c9396 100644
--- a/deepspeech/utils/layer_tools.py
+++ b/deepspeech/utils/layer_tools.py
@@ -33,7 +33,7 @@ def summary(layer: nn.Layer, print_func=print):
     if print_func:
         num_elements = num_elements / 1024**2
         print_func(
-            f"Total parameters: {num_params}, {num_elements:.4f}M elements.")
+            f"Total parameters: {num_params}, {num_elements:.2f} M elements.")
 
 
 def print_grads(model, print_func=print):
@@ -57,7 +57,7 @@ def print_params(model, print_func=print):
             print_func(msg)
     if print_func:
         total = total / 1024**2
-        print_func(f"Total parameters: {num_params}, {total:.4f}M elements.")
+        print_func(f"Total parameters: {num_params}, {total:.2f} M elements.")
 
 
 def gradient_norm(layer: nn.Layer):
diff --git a/examples/tiny/s1/local/export.sh b/examples/tiny/s1/local/export.sh
index b83a13a98..fb0c3cfae 100755
--- a/examples/tiny/s1/local/export.sh
+++ b/examples/tiny/s1/local/export.sh
@@ -5,14 +5,24 @@ if [ $# != 3 ];then
     exit -1
 fi
 
+ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
+echo "using $ngpu gpus..."
+
 config_path=$1
 ckpt_path_prefix=$2
 jit_model_export_path=$3
 
+device=gpu
+if [ ngpu != 0 ];then
+    device=cpu
+fi
+
 python3 -u ${BIN_DIR}/export.py \
+--device ${device} \
+--nproc ${ngpu} \
 --config ${config_path} \
 --checkpoint_path ${ckpt_path_prefix} \
---export_path ${jit_model_export_path} 
+--export_path ${jit_model_export_path}
 
 
 if [ $? -ne 0 ]; then
diff --git a/examples/tiny/s1/local/train.sh b/examples/tiny/s1/local/train.sh
index 3ed533808..47645d4b5 100755
--- a/examples/tiny/s1/local/train.sh
+++ b/examples/tiny/s1/local/train.sh
@@ -10,6 +10,7 @@ echo "using $ngpu gpus..."
 
 config_path=$1
 ckpt_name=$2
+
 device=gpu
 if [ ngpu != 0 ];then
     device=cpu
diff --git a/examples/tiny/s1/run.sh b/examples/tiny/s1/run.sh
index 3b0da66ca..71cd45349 100644
--- a/examples/tiny/s1/run.sh
+++ b/examples/tiny/s1/run.sh
@@ -5,16 +5,16 @@ source path.sh
 source ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
 
 # prepare data
-bash ./local/data.sh
+bash ./local/data.sh || exit -1
 
 # train model, all `ckpt` under `exp` dir
-CUDA_VISIBLE_DEVICES=0 ./local/train.sh conf/conformer.yaml test
-
-# test ckpt 1
-CUDA_VISIBLE_DEVICES=0 ./local/test.sh conf/conformer.yaml exp/test/checkpoints/1
+CUDA_VISIBLE_DEVICES=0 ./local/train.sh conf/conformer.yaml test || exit -1
 
 # avg 1 best model
 ./local/avg.sh exp/test/checkpoints 1
 
+# test ckpt 1
+CUDA_VISIBLE_DEVICES=0 ./local/test.sh conf/conformer.yaml exp/test/checkpoints/avg_1 || exit -1
+
 # export ckpt 1
-./local/export.sh conf/conformer.yaml exp/test/checkpoints/1 exp/test/checkpoints/1.jit.model
\ No newline at end of file
+CUDA_VISIBLE_DEVICES= ./local/export.sh conf/conformer.yaml exp/test/checkpoints/avg_1 exp/test/checkpoints/avg_1.jit.model
diff --git a/examples/tiny/s1/train.profile b/examples/tiny/s1/train.profile
deleted file mode 100644
index 08a9a8684..000000000
Binary files a/examples/tiny/s1/train.profile and /dev/null differ