diff --git a/vllm/compilation/pass_manager.py b/vllm/compilation/pass_manager.py index e07e52be9f..1b1cbe4fa1 100644 --- a/vllm/compilation/pass_manager.py +++ b/vllm/compilation/pass_manager.py @@ -8,13 +8,13 @@ from vllm.logger import init_logger from vllm.platforms import current_platform if current_platform.is_cuda_alike(): + from .activation_quant_fusion import ActivationQuantFusionPass from .fusion import FusionPass from .fusion_attn import AttnFusionPass if current_platform.is_cuda(): from .collective_fusion import AllReduceFusionPass, AsyncTPPass -from .activation_quant_fusion import ActivationQuantFusionPass from .fix_functionalization import FixFunctionalizationPass from .inductor_pass import CustomGraphPass, InductorPass, get_pass_context from .noop_elimination import NoOpEliminationPass