[Bugfix] Fix test_cutlass_moe.py (#44380)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-06-04 14:18:52 -04:00
committed by GitHub
parent 8d9536a775
commit 439203d32c
3 changed files with 20 additions and 9 deletions
+6 -2
View File
@@ -49,10 +49,12 @@ def shuffle_weight(w: torch.Tensor) -> torch.Tensor:
def make_dummy_moe_config(
num_experts: int = 1,
num_local_experts: int | None = None,
experts_per_token: int = 1,
hidden_dim: int = 1,
intermediate_size_per_partition: int = 1,
in_dtype: torch.dtype = torch.bfloat16,
max_num_tokens: int = 512,
) -> FusedMoEConfig:
"""
This is a dummy config for the mk constructor interface
@@ -66,14 +68,16 @@ def make_dummy_moe_config(
experts_per_token=experts_per_token,
hidden_dim=hidden_dim,
intermediate_size_per_partition=intermediate_size_per_partition,
num_local_experts=num_experts,
num_local_experts=num_local_experts
if num_local_experts is not None
else num_experts,
num_logical_experts=num_experts,
moe_parallel_config=FusedMoEParallelConfig.make_no_parallel(),
activation=MoEActivation.SILU,
in_dtype=in_dtype,
device="cuda",
routing_method=RoutingMethodType.TopK,
max_num_tokens=512,
max_num_tokens=max_num_tokens,
)