Add THD-format per-rank batch fetch with padding mask for sequence packing - #6589

Closed

ilml wants to merge 2 commits into

NVIDIA:pull-request/5903from

ilml:split/3386-04-thd-batch-fetch

Commits on Aug 10, 2026

Add THD-format per-rank batch fetch with padding mask for sequence packing

ilml
and
claude
committed
ba18123
Address review: drop the TE wrapper and reuse broadcast_scalars

ilml
and
claude
committed
956a101