Ethanfel e7e7c1cb5a Fix sparse attention mask tiling for temporal windows ...

The local_attn_mask was not being tiled across temporal dimensions,
causing assertion errors in streaming mode and wrong masks otherwise.
Match naxci1 reference: 4D tile/rearrange for Q/K temporal windows,
chunk-based score computation, and topk<=0 guard.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-02-13 18:50:40 +01:00

37 KiB

Raw Blame History

View Raw