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
[ROCm][Bugfix] Compilation passes fix (#22202)
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Gregory Shtrasberg committed
Aug 4, 2025 at 22:12 UTC
031ca762d7bdb566917c8aa39a0294fea89c55ed
1 file changed
+4
-2
vllm/compilation/pass_manager.py
+4
-2
@@ -7,11 +7,13 @@ from vllm.config import VllmConfig
7
from vllm.logger import init_logger
8
from vllm.platforms import current_platform
9
10
-if current_platform.is_cuda():
10
+if current_platform.is_cuda_alike():
11
from .fusion import FusionPass
12
- from .collective_fusion import AllReduceFusionPass, AsyncTPPass
12
from .fusion_attn import AttnFusionPass
13
14
+if current_platform.is_cuda():
15
+ from .collective_fusion import AllReduceFusionPass, AsyncTPPass
16
+
17
from .activation_quant_fusion import ActivationQuantFusionPass
18
from .fix_functionalization import FixFunctionalizationPass
19
from .inductor_pass import CustomGraphPass, InductorPass, get_pass_context