[MoE Refactor] FusedMoE/MoERunner inversion refactor (#41184)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-06-08 10:42:58 -04:00
committed by GitHub
parent 753e9d55e6
commit dc68bd8c41
90 changed files with 2738 additions and 2031 deletions
+2 -2
View File
@@ -146,8 +146,8 @@ def test_quark_int8_w8a8_moe(vllm_runner, tp):
layer = model.model.layers[0]
# MoE experts should use QuarkW8A8Int8MoEMethod
moe = layer.mlp.experts
assert isinstance(moe.quant_method, QuarkW8A8Int8MoEMethod), (
f"Expected QuarkW8A8Int8MoEMethod, got {type(moe.quant_method)}"
assert isinstance(moe._quant_method, QuarkW8A8Int8MoEMethod), (
f"Expected QuarkW8A8Int8MoEMethod, got {type(moe._quant_method)}"
)
# Non-MoE linear layers should use QuarkW8A8Int8
qkv_proj = layer.self_attn.qkv_proj