fix: padding 默认值改为 0，避免 loss_mask 误计 padding 位置

2026-05-15 16:59:42 +08:00
parent cb6bfcb976
commit 625695fd71
2 changed files with 2 additions and 2 deletions
@@ -83,7 +83,7 @@ def cache_jsonl(
    processor: BaseProcessor,
    *,
    pack_size: int = -1,
-    pad_value: int = 1,
+    pad_value: int = 0,
 ) -> List[str]:
    """Tokenize JSONL files and pack them into HDF5 storage.