diff --git a/open_lm/datapreprocess/ray/tokenize_shuffle.py b/open_lm/datapreprocess/ray/tokenize_shuffle.py
index 79bdea79..86a3c959 100644
--- a/open_lm/datapreprocess/ray/tokenize_shuffle.py
+++ b/open_lm/datapreprocess/ray/tokenize_shuffle.py
@@ -234,7 +234,7 @@ def _flush_buffer(self, folder, counter):
                 tokens = [int(x) for x in self.buffer[i]["tokens"]]
                 token_count += len(tokens)
                 json_string = json.dumps(tokens)
-                uid = f"{tar_index_str}_{i:0{digits}}"
+                uid = hashlib.md5(json_string.encode()).hexdigest()
                 sample = {"__key__": uid, "json.gz": json_string}
                 sink.write(sample)
         bio.seek(0)