[Kernel][Misc] Use TORCH_LIBRARY instead of PYBIND11_MODULE for custom ops (#5047)
This commit is contained in:
@@ -8,7 +8,6 @@ import torch
|
||||
import triton
|
||||
import triton.language as tl
|
||||
|
||||
import vllm._moe_C as moe_kernels
|
||||
from vllm import _custom_ops as ops
|
||||
from vllm.logger import init_logger
|
||||
|
||||
@@ -355,7 +354,7 @@ def fused_topk(
|
||||
topk,
|
||||
dtype=torch.int32,
|
||||
device=hidden_states.device)
|
||||
moe_kernels.topk_softmax(
|
||||
ops.topk_softmax(
|
||||
topk_weights,
|
||||
topk_ids,
|
||||
token_expert_indicies,
|
||||
|
||||
Reference in New Issue
Block a user