[Refactor] Remove unused cutlass moe problem size function (#32047)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
@@ -1044,41 +1044,6 @@ def get_cutlass_moe_mm_data(
|
||||
)
|
||||
|
||||
|
||||
def get_cutlass_moe_mm_problem_sizes(
|
||||
topk_ids: torch.Tensor,
|
||||
problem_sizes1: torch.Tensor,
|
||||
problem_sizes2: torch.Tensor,
|
||||
num_experts: int,
|
||||
n: int,
|
||||
k: int,
|
||||
blockscale_offsets: torch.Tensor | None = None,
|
||||
force_swap_ab: bool | None = None,
|
||||
):
|
||||
"""
|
||||
Compute only the per-expert problem sizes needed by the two grouped matrix
|
||||
multiplications used in CUTLASS-based fused MoE.
|
||||
|
||||
The function takes in topk_ids (token→expert mapping) and computes:
|
||||
- problem_sizes1, problem_sizes2: M×N×K sizes of each expert's
|
||||
multiplication for the two grouped MMs
|
||||
used in the fused MoE operation.
|
||||
Optional:
|
||||
- force_swap_ab: If set to True or False, explicitly enable or disable the
|
||||
A/B input swap optimization. If None (default), the swap
|
||||
is selected automatically based on tensor sizes.
|
||||
"""
|
||||
return torch.ops._C.get_cutlass_moe_mm_problem_sizes(
|
||||
topk_ids,
|
||||
problem_sizes1,
|
||||
problem_sizes2,
|
||||
num_experts,
|
||||
n,
|
||||
k,
|
||||
blockscale_offsets,
|
||||
force_swap_ab,
|
||||
)
|
||||
|
||||
|
||||
def get_cutlass_moe_mm_problem_sizes_from_expert_offsets(
|
||||
expert_first_token_offset: torch.Tensor,
|
||||
problem_sizes1: torch.Tensor,
|
||||
|
||||
Reference in New Issue
Block a user