Added inter document masking for manual and flash attention.#434

Open

BlueCrescent wants to merge 17 commits intomainfrom

inter_document_masking_for_attention

Commits on Feb 23, 2026

feat(attention): Added inter document masking for manual and flash attention.
BlueCrescent
committed
feat(data loading): GPT2LLMCollateFn can now determine the sub sequence lengths required for inter document attention masking.
BlueCrescent
committed

Commits on Feb 25, 2026

test(attention): fixed collator tests and improved collator config validators
BlueCrescent
committed

Commits on Feb 26, 2026

fix: directly use correct device + dtype for eos positions extensions

BlueCrescent
and
Copilot
authored

Commits on Feb 27, 2026

chore: remove comment

BlueCrescent
and
rrutmann
authored

Commits on Feb 28, 2026