diff --git a/convlab2/base_models/bert/train_bio.sh b/convlab2/base_models/bert/train_bio.sh
index 59973634fe819c0c2d8b8c7fb6b2b79d420ecc2d..4d6a4d8df84f5811484f8a5459cdd02d20759b4a 100644
--- a/convlab2/base_models/bert/train_bio.sh
+++ b/convlab2/base_models/bert/train_bio.sh
@@ -20,6 +20,7 @@ num_train_epochs=1
 metric_for_best_model="f1"
 
 names=$(echo ${dataset_name} | tr "+" "\n")
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 for name in ${names};
 do
diff --git a/convlab2/base_models/gpt/keyword_extraction/gen_pretraining_data.sh b/convlab2/base_models/gpt/keyword_extraction/gen_pretraining_data.sh
index 8a4290c3f2cd4fcae432d7fb94e4da25ec4ec727..c48f49b52947ff56b68710265d7aae79856c005a 100644
--- a/convlab2/base_models/gpt/keyword_extraction/gen_pretraining_data.sh
+++ b/convlab2/base_models/gpt/keyword_extraction/gen_pretraining_data.sh
@@ -2,6 +2,7 @@ dataset_name="metalwoz+sgd+tm1+tm2+tm3"
 names=$(echo ${dataset_name} | tr "+" "\n")
 model_type="gpt"
 data_dir=data/key2gen_shuffle_noisy/${model_type}/${name}/${dataset_name}
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 train_file="${data_dir}/train.json"
 validation_file="${data_dir}/validation.json"
diff --git a/convlab2/base_models/t5/dst/run_dst_pretrain.sh b/convlab2/base_models/t5/dst/run_dst_pretrain.sh
index f1c5c3d48799a51b208bde12938e15a08d4632a4..119959431f07485f613de28e5f338b307c9647d6 100644
--- a/convlab2/base_models/t5/dst/run_dst_pretrain.sh
+++ b/convlab2/base_models/t5/dst/run_dst_pretrain.sh
@@ -25,6 +25,7 @@ lr=1e-3
 num_train_epochs=1
 
 names=$(echo ${dataset_name} | tr "+" "\n")
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 for name in ${names};
 do
diff --git a/convlab2/base_models/t5/nlg/run_nlg_pretrain.sh b/convlab2/base_models/t5/nlg/run_nlg_pretrain.sh
index 4ff752b64b321955fad5564353002aeaa79b2f30..4d2b440ece35fd448277a4a50b5e3c122606a8ae 100644
--- a/convlab2/base_models/t5/nlg/run_nlg_pretrain.sh
+++ b/convlab2/base_models/t5/nlg/run_nlg_pretrain.sh
@@ -25,6 +25,7 @@ lr=1e-3
 num_train_epochs=1
 
 names=$(echo ${dataset_name} | tr "+" "\n")
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 for name in ${names};
 do
diff --git a/convlab2/base_models/t5/nlu/run_nlu_pretrain.sh b/convlab2/base_models/t5/nlu/run_nlu_pretrain.sh
index 3a059c2d27c809a3a1cf26b3a49160cab96dd1d4..ccc7c08fcf1ebbd5d71ed800ce56b7413432631a 100644
--- a/convlab2/base_models/t5/nlu/run_nlu_pretrain.sh
+++ b/convlab2/base_models/t5/nlu/run_nlu_pretrain.sh
@@ -25,6 +25,7 @@ lr=1e-3
 num_train_epochs=1
 
 names=$(echo ${dataset_name} | tr "+" "\n")
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 for name in ${names};
 do
diff --git a/convlab2/base_models/t5/rg/run_rg.sh b/convlab2/base_models/t5/rg/run_rg.sh
index ac065b81fefa2c406a469a6658263477f1fbe18d..308c639c71ceeb8957188ab499341e8fa2229943 100644
--- a/convlab2/base_models/t5/rg/run_rg.sh
+++ b/convlab2/base_models/t5/rg/run_rg.sh
@@ -23,6 +23,7 @@ lr=1e-3
 num_train_epochs=1
 
 names=$(echo ${dataset_name} | tr "+" "\n")
+rm -r ${data_dir}
 mkdir -p ${data_dir}
 for name in ${names};
 do