diff --git a/examples/mcore/llama32/data_convert_llama32_ppo.sh b/examples/mcore/llama32/data_convert_llama32_ppo.sh
new file mode 100644
index 0000000000000000000000000000000000000000..9e8e4b097f16f1d83dfa30e321f92fbfefc4d433
--- /dev/null
+++ b/examples/mcore/llama32/data_convert_llama32_ppo.sh
@@ -0,0 +1,14 @@
+# 请按照您的真实环境修改 set_env.sh 路径
+source /usr/local/Ascend/ascend-toolkit/set_env.sh
+mkdir ./dataset/llama32-hf/
+
+python ./preprocess_data.py \
+    --input ./dataset/descriptiveness-00000-of-00001.parquet \
+    --tokenizer-name-or-path ./model_from_hf/llama32-hf/ \
+    --output-prefix ./dataset/llama32-hf/descriptiveness \
+    --workers 16 \
+    --log-interval 1000 \
+    --tokenizer-type PretrainedFromHF \
+    --handler-name  PPOAlpacaStyleInstructionHandler \
+    --prompt-type llama3 \
+    --map-keys '{"prompt":"prompt", "query":"", "response": "prompt", "system":""}'
\ No newline at end of file