diff --git a/chinese-instruct.py b/chinese-instruct.py
new file mode 100644
index 0000000..9e17967
--- /dev/null
+++ b/chinese-instruct.py
@@ -0,0 +1,34 @@
+# Mxode/Chinese-Reasoning-Distil-Data
+
+from datasets import DatasetDict
+from datasets import load_dataset, concatenate_datasets
+from utils import process_dataset
+
+
+def build_prompt(query, response) -> str:
+    return f"<|user|> {query} <|system|> <bos>{response}<eos>\n"
+
+
+def process_func(input_dict: dict):
+    query = input_dict["prompt"]
+    response = input_dict["response"]
+    return build_prompt(query, response)
+
+
+if __name__ == "__main__":
+    all_data = ['stem_zh', 'infinity-instruct', 'firefly', 'magpie', 'dpsk-r1-distil', 
+                'coig-cqia', 'disc-law', 'neo_sft_phase2', 'chinese-medical', 'chinese-reasoning-distil', 
+                'psycho-10k-dpsk-r1', 'sof-c-zh', 'industryinstruction', 'Chinese-QA-AFAF']
+    
+    datasets = []
+    for subset in all_data:
+        ds = load_dataset("Mxode/Chinese-Instruct", name=subset)
+        datasets.append(ds["train"]) 
+    
+    combined_dataset = concatenate_datasets(datasets)
+    
+    process_dataset(
+        dataset_dict=DatasetDict({"train": combined_dataset}),
+        output_subdir="chinese-instruct",
+        process_func=process_func,
+    )
\ No newline at end of file
diff --git a/utils.py b/utils.py
index f4701ab..392a507 100644
--- a/utils.py
+++ b/utils.py
@@ -76,6 +76,7 @@ def dump_pkl_files(
                         cur_tensor = F.pad(
                             cur_tensor, 
                             (0, packing_size - cur_tensor.numel()),
+                            'constant',
                             tokenizer.pad_id
                         )
                         packages.append(cur_tensor)