Amir Ghamarian db98d0e5aa Use direct sweep_tile in decode pipeline. ...

Replace the local tile_sweeper wrapper call with sweep_tile(...) to keep the decode path simpler while preserving behavior.

Made-with: Cursor

2026-03-30 10:24:09 +00:00

..

kernel

Add fused topk_softmax_decode kernel for M=1 MoE decode

2026-03-29 18:06:03 +00:00

pipeline

Use direct sweep_tile in decode pipeline.

2026-03-30 10:24:09 +00:00