ViperEkura 4e508afa2d fix : SFT pipeline position_ids default & doc boundary preservation ...

- change position_ids_mode default from "none" to "doc_reset" so SFT preprocessing always generates position_ids (was causing dataset load KeyError)
- generate per-doc position_ids before packing (doc_reset mode), preserving document boundaries for BFD packing (cross-doc attention leak fix)
- change _align_bucket padding from [1] to [0] to avoid accidentally training on loss_mask padding

2026-07-04 15:59:11 +08:00

..

architecture.md

fix: align docs with actual code (40+ inconsistencies)

2026-06-30 20:47:23 +08:00

dataflow.md

fix: align docs with actual code (40+ inconsistencies)

2026-06-30 20:47:23 +08:00

inference.md

fix: align docs with actual code (40+ inconsistencies)

2026-06-30 20:47:23 +08:00

params.md

fix: align docs with actual code (40+ inconsistencies)

2026-06-30 20:47:23 +08:00

preprocessing.md

fix : SFT pipeline position_ids default & doc boundary preservation

2026-07-04 15:59:11 +08:00

README-zh-CN.md

docs: 重构 README 结构，全文档添加目录导航

2026-06-19 13:53:22 +08:00

training.md

fix: align docs with actual code (40+ inconsistencies)

2026-06-30 20:47:23 +08:00