refactor : Pipeline 去除去重,ids 重命名为 sequence,泛型透传

- 移除 Pipeline 内置去重逻辑及 dedup_signature 工具函数
- 删除 ProcessingConfig.deduplicate 字段
- builder 返回 'sequence' 替代 'ids',与 dataset 层统一
- pipeline 纯透传,泛型处理任意 key 补齐默认值
This commit is contained in:
2026-05-31 15:14:27 +08:00
parent 14f83cbdac
commit 01ce1fb9e3
6 changed files with 32 additions and 52 deletions
+1 -1
View File
@@ -151,7 +151,7 @@ class SectionedMaskBuilder(BaseMaskBuilder):
return None
result: dict = {
"ids": all_ids,
"sequence": all_ids,
"domain": _extract_domain(item, config.output.domain_key),
}
if not all(m == 1 for m in loss_mask):