[KDA] KDA MTP decode: recurrent + KVBuffer chunkwise verify + flush - #96

Merged

Longxmas merged 19 commits into

inclusionAI:mainfrom

Longxmas:feat/kda-mtp-verify-4ops

Jul 9, 2026

Commits on Jun 16, 2026

[KDA] KDA MTP decode recurrent
Longxmas
committed
1f2a76c
[KDA] KDA MTP decode kvbuffer chunkwise verify + flush
Longxmas
committed
7008d27
[KDA] MTP decode review fixes: per-request runtime accept-len, skip padded CTAs, empty dummies
Longxmas
committed
c017db1

Commits on Jun 23, 2026

[KDA] KDA MTP decode kvbuffer cg verify: bank-conflict fix + 3xTF32
Longxmas
committed
9bf0493

Commits on Jun 24, 2026

[KDA] KDA MTP decode: lower_bound safe gate for vk/ws/kv/cg/tp gates
Longxmas
committed
cf3c1e7
[KDA] KDA MTP decode: dynamic-N batch dim + ruff lint fixes
Longxmas
committed
5fbb826
[chore] apply ruff format
Longxmas
committed
71d73f2

Commits on Jun 25, 2026

[KDA] KDA MTP decode kvbuffer: batched all-layers flush
Longxmas
committed
0da49c7

Commits on Jul 2, 2026

Commits on Jul 3, 2026

Commits on Jul 5, 2026

[kda-mtp] recurrent_ws + kvbuffer verify: accept K-contiguous strided q/k/v (dyn-stride)
Longxmas
committed
d95fb16

Commits on Jul 6, 2026

Commits on Jul 7, 2026

[kda-mtp] recurrent dispatch: route T==4 large-batch to recurrent_ws
Longxmas
committed
9e69e4e

Commits on Jul 8, 2026

Merge upstream main into feat/kda-mtp-verify-4ops
Longxmas
committed
dcad512