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 13, 02:00 UTC
[XPU] work around for sp, avoid custom op import error (#28822)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Kunshang Ji committed
Nov 17, 2025 at 21:16 UTC
1b82fb0ad3cea2e1a31da4fa20dd736a8a181089
1 file changed
+1
-1
vllm/compilation/pass_manager.py
+1
-1
@@ -18,6 +18,7 @@ if current_platform.is_cuda_alike():
18
from .fusion import RMSNormQuantFusionPass
19
from .fusion_attn import AttnFusionPass
20
from .qk_norm_rope_fusion import QKNormRoPEFusionPass
21
+ from .sequence_parallelism import SequenceParallelismPass
22
23
if current_platform.is_cuda():
24
from .collective_fusion import AllReduceFusionPass, AsyncTPPass
@@ -25,7 +26,6 @@ if current_platform.is_cuda():
26
from .fix_functionalization import FixFunctionalizationPass
27
from .inductor_pass import CustomGraphPass, InductorPass, get_pass_context
28
from .noop_elimination import NoOpEliminationPass
28
-from .sequence_parallelism import SequenceParallelismPass
29
30
logger = init_logger(__name__)
31