[Model][MoE] Load MegaMoE experts directly into the DeepGEMM layout (#60096)

Signed-off-by: aoshen02 <aoshen@inferact.ai>
Signed-off-by: Aaron Hao <hao-aaron@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Aaron Hao <hao-aaron@users.noreply.github.com>
This commit is contained in:
aoshen02
2026-10-13 11:52:08 -07:00
committed by GitHub
co-authored by Claude Opus 5.5 Aaron Hao
parent 624a5ddafd
commit c5bda658d2
11 changed files with 564 additions and 891 deletions
+8 -1
View File
@@ -242,7 +242,14 @@ def test_v41_dspark_loads_linear_scales(
linear_scale_name=scale_name,
pad_shared_expert=False,
model=SimpleNamespace(
layers=[SimpleNamespace(ffn=SimpleNamespace(use_native_mega_moe=False))],
layers=[
SimpleNamespace(
ffn=SimpleNamespace(
use_native_mega_moe=False,
experts=SimpleNamespace(has_fused_shared_experts=False),
)
)
],
confidence_head=None,
),
named_parameters=lambda: [(runtime_name, param)],