[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 -4
View File
@@ -166,12 +166,11 @@ class TestData:
num_experts=e,
experts_per_token=topk,
hidden_dim=k,
intermediate_size_per_partition=n,
intermediate_size=n,
num_local_experts=e,
num_logical_experts=e,
moe_parallel_config=layer.moe_parallel_config,
in_dtype=hidden_states.dtype,
is_act_and_mul=is_gated,
routing_method=layer.routing_method_type,
activation=activation,
device=w13_quantized.device,
@@ -339,14 +338,13 @@ def test_flashinfer_cutlass_moe_fp8_no_graph(
num_experts=e,
experts_per_token=topk,
hidden_dim=k,
intermediate_size_per_partition=n,
intermediate_size=n,
num_local_experts=e,
num_logical_experts=e,
activation=activation,
device="cuda",
moe_parallel_config=FusedMoEParallelConfig.make_no_parallel(),
in_dtype=torch.bfloat16,
is_act_and_mul=activation.is_gated,
routing_method=RoutingMethodType.TopK,
max_num_tokens=next_power_of_2(m),
)