[TRTLLM-15284][feat] add Kimi K3 SiTU MegaMoE support - #17063

Merged

longlee0622 merged 15 commits into

NVIDIA:feat/kimi_k3from

longlee0622:agent/kimi-k3-megamoe-deepgemm-2.6.1

Aug 13, 2026

Commits on Aug 11, 2026

[None][feat] add Kimi K3 SiTU MegaMoE support
longlee0622
committed
3a4152b
[None][chore] map Kimi DeepGEMM fork inputs
longlee0622
committed
e906e6b
[None][fix] Let MegaMoE load the Kimi K3 packed MXFP4 checkpoint

Xin Guan
authored and
longlee0622
committed
63fc93f
[None][test] support MegaMoE packed weight fixture
longlee0622
committed
0ef1d48
[None][test] Move Kimi MoE imports to module scope
longlee0622
committed
d420dbe
[None][fix] honor explicit Kimi MoE backend selection
longlee0622
committed
fb3a9d7
[None][chore] Pin DeepGEMM SiTU support upstream
longlee0622
committed
3ed8bd9
[None][chore] Refresh DeepGEMM attribution checksum
longlee0622
committed
189c715
[TRTLLM-15284][fix] Restrict Kimi fused latent path backend
longlee0622
committed
6e45411
[TRTLLM-15284][fix] Reject unsupported Kimi MoE backends
longlee0622
committed
a26f590
[TRTLLM-15284][fix] Clean up Kimi MegaMoE test group
longlee0622
committed
5784902
[TRTLLM-15284][fix] Propagate Kimi SiTU config explicitly
longlee0622
committed
6022682
[TRTLLM-15284][fix] Scope Kimi MegaMoE token capacity
longlee0622
committed
c976d60
[TRTLLM-15284][test] Fix Kimi MegaMoE test setup
longlee0622
committed
7ffa048

Commits on Aug 12, 2026

[TRTLLM-15284][test] Clear MegaMoE cache in test teardown
longlee0622
committed
936648e