24 lines
435 B
Python
24 lines
435 B
Python
import torch
|
|
|
|
|
|
def moe_align_block_size(
|
|
topk_ids,
|
|
num_experts,
|
|
block_size,
|
|
sorted_token_ids,
|
|
experts_ids,
|
|
num_tokens_post_pad,
|
|
token_cnts_buffer,
|
|
cumsum_buffer,
|
|
):
|
|
torch.ops.sgl_kernel.moe_align_block_size(
|
|
topk_ids,
|
|
num_experts,
|
|
block_size,
|
|
sorted_token_ids,
|
|
experts_ids,
|
|
num_tokens_post_pad,
|
|
token_cnts_buffer,
|
|
cumsum_buffer,
|
|
)
|