biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[refactor] refactor memory constants usage (#31865)

Browse Source 
Signed-off-by: Andy Xie <andy.xning@gmail.com>
This commit is contained in:
Ning Xie
2026-01-08 02:37:31 +08:00
committed by GitHub
parent f347ac6c34
commit c907d22158
9 changed files with 33 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/platforms/cpu.py
View File

@@ -140,6 +140,7 @@ class CpuPlatform(Platform):
@classmethod
def get_device_total_memory(cls, device_id: int = 0) -> int:
from vllm.utils.mem_constants import GiB_bytes
from vllm.utils.mem_utils import format_gib
kv_cache_space = envs.VLLM_CPU_KVCACHE_SPACE
node_dir = "/sys/devices/system/node"
@@ -153,10 +154,9 @@ class CpuPlatform(Platform):
free_cpu_memory = psutil.virtual_memory().total // num_numa_nodes
DEFAULT_CPU_MEM_UTILIZATION = 0.5
kv_cache_space = int(free_cpu_memory * DEFAULT_CPU_MEM_UTILIZATION)
kv_cache_space_gib = kv_cache_space / GiB_bytes
logger.warning_once(
"VLLM_CPU_KVCACHE_SPACE not set. Using "
f"{kv_cache_space_gib:.2f} GiB for KV cache."
"VLLM_CPU_KVCACHE_SPACE not set. Using %s GiB for KV cache.",
format_gib(kv_cache_space),
)
else:
kv_cache_space *= GiB_bytes