fix: cache_h5 移除重复 --batch-size 参数，默认输出格式改为 bin

scripts/cache_h5.py

@@ -66,12 +66,6 @@ def main():
         default=1_000,
         help="Merge every N packed chunks into one tensor, <=0 to disable (default: 1000)",
     )
-    parser.add_argument(
-        "--batch-size",
-        type=int,
-        default=256,
-        help="Records tokenized per batch (default: 256)",
-    )
     parser.add_argument(
         "--log-level",
         default="INFO",
@@ -87,9 +81,9 @@ def main():
     parser.add_argument(
         "-f",
         "--output-format",
-        default="h5",
+        default="bin",
         choices=["h5", "bin"],
-        help="Output format: h5 or bin (default: h5)",
+        help="Output format: h5 or bin (default: bin)",
     )
     args = parser.parse_args()