A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Platform] import activation_quant_fusion for CUDA only (#23882)
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
wangxiyuan committed
Aug 29, 2025 at 13:54 UTC
6597d7a4566d344835f5a90621397d8fee490b10
1 file changed
+1
-1
vllm/compilation/pass_manager.py
+1
-1
@@ -8,13 +8,13 @@ from vllm.logger import init_logger
8
from vllm.platforms import current_platform
9
10
if current_platform.is_cuda_alike():
11
+ from .activation_quant_fusion import ActivationQuantFusionPass
12
from .fusion import FusionPass
13
from .fusion_attn import AttnFusionPass
14
15
if current_platform.is_cuda():
16
from .collective_fusion import AllReduceFusionPass, AsyncTPPass
17
17
-from .activation_quant_fusion import ActivationQuantFusionPass
18
from .fix_functionalization import FixFunctionalizationPass
19
from .inductor_pass import CustomGraphPass, InductorPass, get_pass_context
20
from .noop_elimination import NoOpEliminationPass