Fix fused_moe import for non-NPU devices (#25076)
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
This commit is contained in:
co-authored by
Brayden Zhong
parent
00d8730d05
commit
52d4c697bb
@@ -62,9 +62,7 @@ from sglang.srt.utils import add_prefix, is_npu
|
|||||||
|
|
||||||
_is_npu = is_npu()
|
_is_npu = is_npu()
|
||||||
|
|
||||||
if not _is_npu:
|
if _is_npu:
|
||||||
from sglang.srt.layers.moe.fused_moe_triton import fused_moe
|
|
||||||
else:
|
|
||||||
from sglang.srt.hardware_backend.npu.quantization.fused_moe_method_npu import (
|
from sglang.srt.hardware_backend.npu.quantization.fused_moe_method_npu import (
|
||||||
fused_moe_npu as fused_moe,
|
fused_moe_npu as fused_moe,
|
||||||
)
|
)
|
||||||
|
|||||||
Reference in New Issue
Block a user