[None][feat] bring up Kimi K3 NVFP4 with CUTLASS and cuteDSL MegaMoE SiTU - #17865

Merged

longlee0622 merged 9 commits into

NVIDIA:mainfrom

longlee0622:agent/kimi-k3-nvfp4-megamoe-situ

Aug 21, 2026

Commits on Aug 20, 2026

[None][fix] Add SiTu to the Python gated-activation list
longlee0622
committed
ce94087
[None][fix] Record the checkpoint directory on lazily-loaded weights
longlee0622
committed
dabdb41
[None][feat] Kimi K3: serve the nvidia/Kimi-K3-NVFP4 checkpoint
longlee0622
committed
57e11fc
[None][feat] Kimi K3: MegaMoE CuteDSL SiTU support for the NVFP4 checkpoint
longlee0622
committed
698f8fe
[None][fix] Enable Kimi K3 SiTU on CUTLASS
longlee0622
committed
d622b63
Remove local doc reference from MegaMoE comment
longlee0622
committed
e9d5e7c
Align Kimi FP8 checkpoint stash gate
longlee0622
committed
6b09ef0
Address Kimi SiTu review feedback
longlee0622
committed
08abae0
[None][fix] Kimi K3: size CUTLASS SiTU alpha/beta with the shared EP partition
longlee0622
committed
d45ca60