File tree Expand file tree Collapse file tree 1 file changed +5
-2
lines changed
vllm/model_executor/layers/fused_moe Expand file tree Collapse file tree 1 file changed +5
-2
lines changed Original file line number Diff line number Diff line change @@ -792,9 +792,12 @@ def __init__(
792
792
if quant_config and quant_config .get_name () == "mxfp4" :
793
793
from vllm .model_executor .layers .quantization .mxfp4 import ( # noqa: E501
794
794
should_use_flashinfer_mxfp4 , should_use_flashinfer_mxfp4_bf16 )
795
- if current_platform .is_rocm () or (should_use_flashinfer_mxfp4 () and current_platform .is_device_capability (100 )):
795
+ if current_platform .is_rocm () or (
796
+ should_use_flashinfer_mxfp4 ()
797
+ and current_platform .is_device_capability (100 )):
796
798
hidden_size = round_up (hidden_size , 256 )
797
- elif should_use_flashinfer_mxfp4_bf16 () and current_platform .is_device_capability (90 ):
799
+ elif should_use_flashinfer_mxfp4_bf16 (
800
+ ) and current_platform .is_device_capability (90 ):
798
801
hidden_size = round_up (hidden_size , 128 )
799
802
800
803
# For smuggling this layer into the fused moe custom op
You can’t perform that action at this time.
0 commit comments