refactor : 统一 SectionedMaskBuilder,支持可配置 dtype

- 三合一 MaskBuilder,移除 chat/instruction/text,统一为 sections 配置
- OutputConfig 增加 dtype 字段 (per-key,默认 int32)
- 移除 from __future__ import annotations
- 测试适配新配置格式
This commit is contained in:
2026-05-31 14:24:10 +08:00
parent 2a65c3314c
commit dbe5891201
5 changed files with 330 additions and 206 deletions
+2 -6
View File
@@ -1,18 +1,14 @@
from astrai.preprocessing.builder import (
BaseMaskBuilder,
ChatMaskBuilder,
InstructionMaskBuilder,
MaskBuilderFactory,
TextMaskBuilder,
SectionedMaskBuilder,
)
from astrai.preprocessing.pipeline import Pipeline, dedup_signature, filter_by_length
__all__ = [
"BaseMaskBuilder",
"ChatMaskBuilder",
"InstructionMaskBuilder",
"MaskBuilderFactory",
"TextMaskBuilder",
"SectionedMaskBuilder",
"Pipeline",
"dedup_signature",
"filter_by_length",