[GPTOSS][DP/EP][Marlin] Enable GPTOSS Batched DP/EP using Marlin kernels (#25997)
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
This commit is contained in:
co-authored by
Varun Sundar Rabindranath
parent
2ed8b6b3d0
commit
fb0571b077
@@ -1789,6 +1789,24 @@ def moe_align_block_size(
|
||||
)
|
||||
|
||||
|
||||
def batched_moe_align_block_size(
|
||||
max_tokens_per_batch: int,
|
||||
block_size: int,
|
||||
expert_num_tokens: torch.Tensor,
|
||||
sorted_ids: torch.Tensor,
|
||||
expert_ids: torch.Tensor,
|
||||
num_tokens_post_pad: torch.Tensor,
|
||||
) -> None:
|
||||
torch.ops._moe_C.batched_moe_align_block_size(
|
||||
max_tokens_per_batch,
|
||||
block_size,
|
||||
expert_num_tokens,
|
||||
sorted_ids,
|
||||
expert_ids,
|
||||
num_tokens_post_pad,
|
||||
)
|
||||
|
||||
|
||||
def moe_wna16_gemm(
|
||||
input: torch.Tensor,
|
||||
output: torch.Tensor,
|
||||
|
||||
Reference in New Issue
Block a user