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 5, 01:00 UTC
[ZenCPU] AMD Zen CPU Backend with supported dtypes via zentorch weekly (#39967)
Signed-off-by: Chinmay Kulkarni <Chinmay.Kulkarni@amd.com>
Chinmay-Kulkarni-AMD committed
Apr 18, 2026 at 11:52 UTC
87518c302797be0b20bd720fb3a7e195a517bc66
2 files changed
+11
-1
setup.py
+3
-1
@@ -1085,7 +1085,9 @@ setup(
1085
install_requires=get_requirements(),
1086
extras_require={
1087
# AMD Zen CPU optimizations via zentorch
1088
- "zen": ["zentorch"],
1088
+ "zen": [
1089
+ "zentorch-weekly==5.2.1.dev20260408"
1090
+ ], # Zentorch has weekly releases. This pulls the known-good version.
1091
"bench": ["pandas", "matplotlib", "seaborn", "datasets", "scipy", "plotly"],
1092
"tensorizer": ["tensorizer==2.10.1"],
1093
"fastsafetensors": ["fastsafetensors >= 0.2.2"],
vllm/platforms/zen_cpu.py
+8
@@ -1,6 +1,8 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
+import torch
5
+
6
from vllm.logger import init_logger
7
from vllm.platforms.cpu import CpuPlatform
8
@@ -22,3 +24,9 @@ class ZenCpuPlatform(CpuPlatform):
24
def is_zen_cpu(self) -> bool:
25
# is_cpu() also returns True for this platform (inherited from CpuPlatform).
26
return True
27
+
28
+ # Currently, AMD CPUs do not support float16 compute.
29
+ # Hence explicitly return bfloat16 and float32.
30
+ @property
31
+ def supported_dtypes(self) -> list[torch.dtype]:
32
+ return [torch.bfloat16, torch.float32]