[SM90] Validate MegaMoE symmetric buffer peer mappings - #79

Open

Andrew88Lin wants to merge 24 commits into

sgl-project:devfrom

Andrew88Lin:fix/sm90-megamoe-nvlink-domain-check

Commits on Jun 27, 2026

Support tvm-ffi interfaces

Fridge003
and
rainj-me
committed
88db27f
Support Wheel Compilation
Fridge003
committed
495fcfd
Export the PDL utils of DeepGEMM (#34)

b8zhong
authored and
Fridge003
committed
3036815
Expose BF16 grouped GEMM wrappers

popsiclexu
authored and
Fridge003
committed
607503a
Setup tests for sgl-deep-gemm

Fridge003
and
b8zhong
committed
00c42b8
Sm90 mega moe on sgl dev (#36)

authored and
Fridge003
committed
a8459f1
Add Hopper mega moe test to runner (#46)
Fridge003
committed
eed753c
chore: bump apache-tvm-ffi 0.1.9 -> 0.1.11 (#47)

MartinHua
authored and
Fridge003
committed
f3d0ca5
Add FP4 acts + MXF4 kind support and fused mega_moe_pre_dispatch kernel (#27)

authored and
Fridge003
committed
bb569e2
Add DG_USE_FP8_COMBINE: FP8 + per-row UE8M0 SF on the second a2a (combine path) (#28)

authored and
Fridge003
committed
b111331
Add tvm-ffi wrapper for w4a4 megamoe
Fridge003
committed
d8f874c

Commits on Jun 29, 2026

Rebase and fix
Fridge003
committed
731e7c7

Commits on Jul 9, 2026

Raise DeepGEMM barrier timeout to 300 seconds (#57)
weireweire
authored
07ffd88
[SM90] Optimize FP8 MegaMoE small-batch decode path with swapAB (#48)

qiushixiaoyu
and
yinding
authored
c28bb29
Fix a race condition where in-flight tensormaps are updated in-place (#343)

dfyz
authored and
Fridge003
committed
67138ed

Commits on Jul 14, 2026

feat: add sm120 support for DeepGEMM (#56)

leavelet
and
RayWang96
authored
4bb3467

Commits on Jul 18, 2026

[SM90] Use continuous FP32 activation scale in MegaMoE (drop UE8M0) (#63)

authored
4a25337

Commits on Jul 23, 2026

SM120: MoE grouped GEMM decode optimizations: skip padding I/O, BLOCK_M=32 (#59)

leavelet
authored and
Fridge003
committed
7fbb623
Cover SM120 GPUs in CI test (#61)
Fridge003
committed
d357352
tests: keep tests/ aligned with upstream; move SM120 decode coverage to sgl_deep_gemm/tests (#65)

leavelet
authored and
Fridge003
committed
1a1c9a5

Commits on Jul 29, 2026

Add Kimi-K3 SiTU activation (#67)
Fridge003
authored
75f6062

Commits on Aug 14, 2026

[SM100] Select SiTU through the activation argument (#78)
Fridge003
authored
80b2c44

Commits on Aug 25, 2026

[SM90] Validate MegaMoE symmetric buffer peer mappings
andrewchen_microsoft
committed
fc5e851
[SM90] Reject null peer pointers explicitly
andrewchen_microsoft
committed
9cd49a0