[NPU] Bugfix for qwen35 on NPU (#19756)
This commit is contained in:
@@ -33,11 +33,13 @@ if is_cuda():
|
||||
|
||||
causal_conv1d_fn = causal_conv1d_fn_cuda
|
||||
elif is_npu():
|
||||
from sgl_kernel_npu.fla.fused_gdn_gating import fused_gdn_gating_npu
|
||||
from sgl_kernel_npu.mamba.causal_conv1d import (
|
||||
causal_conv1d_fn_npu,
|
||||
causal_conv1d_update_npu,
|
||||
)
|
||||
|
||||
fused_gdn_gating = fused_gdn_gating_npu
|
||||
causal_conv1d_fn = causal_conv1d_fn_npu
|
||||
causal_conv1d_update = causal_conv1d_update_npu
|
||||
elif is_cpu():
|
||||
|
||||
Reference in New Issue
Block a user