diff --git a/examples/speechcommand_v1/s0/run.sh b/examples/speechcommand_v1/s0/run.sh
index 44bf258..b18a856 100755
--- a/examples/speechcommand_v1/s0/run.sh
+++ b/examples/speechcommand_v1/s0/run.sh
@@ -7,13 +7,13 @@
 export CUDA_VISIBLE_DEVICES="0"
 
 stage=-1
-stop_stage=2
+stop_stage=4
 num_keywords=11
 
 config=conf/mdtc.yaml
 norm_mean=false
 norm_var=false
-gpu_id=4
+gpu_id=0
 
 checkpoint=
 dir=exp/mdtc
@@ -79,3 +79,30 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     $cmvn_opts \
     ${checkpoint:+--checkpoint $checkpoint}
 fi
+
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+  # Do model average
+  python kws/bin/average_model.py \
+    --dst_model $score_checkpoint \
+    --src_path $dir  \
+    --num ${num_average} \
+    --val_best
+
+  # Testing
+  result_dir=$dir/test_$(basename $score_checkpoint)
+  mkdir -p $result_dir
+  python kws/bin/compute_accuracy.py --gpu 3 \
+    --config $dir/config.yaml \
+    --test_data data/test/data.list \
+    --batch_size 256 \
+    --num_workers 8 \
+    --checkpoint $score_checkpoint
+fi
+
+
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+  python kws/bin/export_jit.py --config $dir/config.yaml \
+    --checkpoint $score_checkpoint \
+    --output_file $dir/final.zip \
+    --output_quant_file $dir/final.quant.zip
+fi
diff --git a/kws/bin/compute_accuracy.py b/kws/bin/compute_accuracy.py
new file mode 100644
index 0000000..a6cde27
--- /dev/null
+++ b/kws/bin/compute_accuracy.py
@@ -0,0 +1,102 @@
+# Copyright (c) 2021 Binbin Zhang(binbzha@qq.com)
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import argparse
+import copy
+import logging
+import os
+
+import torch
+import yaml
+from torch.utils.data import DataLoader
+
+from kws.dataset.dataset import Dataset
+from kws.model.kws_model import init_model
+from kws.utils.checkpoint import load_checkpoint
+from kws.utils.executor import Executor
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='recognize with your model')
+    parser.add_argument('--config', required=True, help='config file')
+    parser.add_argument('--test_data', required=True, help='test data file')
+    parser.add_argument('--gpu',
+                        type=int,
+                        default=-1,
+                        help='gpu id for this rank, -1 for cpu')
+    parser.add_argument('--checkpoint', required=True, help='checkpoint model')
+    parser.add_argument('--batch_size',
+                        default=16,
+                        type=int,
+                        help='batch size for inference')
+    parser.add_argument('--num_workers',
+                        default=0,
+                        type=int,
+                        help='num of subprocess workers for reading')
+    parser.add_argument('--pin_memory',
+                        action='store_true',
+                        default=False,
+                        help='Use pinned memory buffers used for reading')
+    parser.add_argument('--prefetch',
+                        default=100,
+                        type=int,
+                        help='prefetch number')
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+    logging.basicConfig(level=logging.DEBUG,
+                        format='%(asctime)s %(levelname)s %(message)s')
+    os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu)
+
+    with open(args.config, 'r') as fin:
+        configs = yaml.load(fin, Loader=yaml.FullLoader)
+
+    test_conf = copy.deepcopy(configs['dataset_conf'])
+    test_conf['filter_conf']['max_length'] = 102400
+    test_conf['filter_conf']['min_length'] = 0
+    test_conf['speed_perturb'] = False
+    test_conf['spec_aug'] = False
+    test_conf['shuffle'] = False
+    test_conf['feature_extraction_conf']['dither'] = 0.0
+    test_conf['batch_conf']['batch_size'] = args.batch_size
+
+    test_dataset = Dataset(args.test_data, test_conf)
+    test_data_loader = DataLoader(test_dataset,
+                                  batch_size=None,
+                                  pin_memory=args.pin_memory,
+                                  num_workers=args.num_workers)
+
+    # Init asr model from configs
+    model = init_model(configs['model'])
+
+    load_checkpoint(model, args.checkpoint)
+    use_cuda = args.gpu >= 0 and torch.cuda.is_available()
+    device = torch.device('cuda' if use_cuda else 'cpu')
+    model = model.to(device)
+    executor = Executor()
+    model.eval()
+    training_config = configs['training_config']
+    with torch.no_grad():
+        test_loss, test_acc = executor.test(model, test_data_loader, device,
+                                            training_config)
+        logging.info('Test Loss {} Acc {}'.format(test_loss, test_acc))
+
+
+if __name__ == '__main__':
+    main()