[KDA] Fuse causal conv1d into MTP verify decode - #107

Merged

Longxmas merged 9 commits into

inclusionAI:mainfrom

Longxmas:feat/kda-conv-mtp-fuse

Aug 4, 2026

Commits on Jul 9, 2026

[kda-conv-mtp] fused causal-conv1d + KDA MTP verify decode (vk + warp-spec ws)
Longxmas
committed
11abdb8

Commits on Jul 15, 2026

[kda-conv-mtp] fused conv+MTP verify: small_batch (2-warp pipelined) + large_batch (warp-spec)
Longxmas
committed
15ff7be

Commits on Jul 16, 2026

[kda-conv-mtp] 0716 microopt: producer v-input scalarization + large_batch store dedup + host cleanups + T-aware dispatch
Longxmas
committed
4c73c94

Commits on Jul 20, 2026

[kda-conv-mtp] fix state indexing and qk writeback race
Longxmas
committed
2288183

Commits on Jul 22, 2026

[kda-conv-mtp] add fused conv KVBuffer verify
Longxmas
committed
5df14b9

Commits on Jul 24, 2026

[kda-mtp] simplify fused conv KVBuffer path
Longxmas
committed
8fee2d4
Merge upstream main into KDA MTP fused branch
Longxmas
committed
a4d271b

Commits on Jul 27, 2026

[kda-mtp] address review feedback and CI checks
Longxmas
committed
ba9986a

Commits on Aug 4, 2026

[kda-mtp] validate recurrent verify input length
Longxmas
committed
ff0bcf0