[cuda] manually import the correct pynvml module (#12679)

fixes problems like https://github.com/vllm-project/vllm/pull/12635 and
https://github.com/vllm-project/vllm/pull/12636 and
https://github.com/vllm-project/vllm/pull/12565

---------

Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2025-02-03 15:58:21 +08:00
committed by GitHub
parent b9986454fe
commit ad4a9dc817
3 changed files with 56 additions and 9 deletions
+2 -1
View File
@@ -33,7 +33,8 @@ def cuda_platform_plugin() -> Optional[str]:
is_cuda = False
try:
import pynvml
from vllm.utils import import_pynvml
pynvml = import_pynvml()
pynvml.nvmlInit()
try:
if pynvml.nvmlDeviceGetCount() > 0:
+2 -8
View File
@@ -8,7 +8,6 @@ from functools import lru_cache, wraps
from typing import (TYPE_CHECKING, Callable, List, Optional, Tuple, TypeVar,
Union)
import pynvml
import torch
from typing_extensions import ParamSpec
@@ -16,6 +15,7 @@ from typing_extensions import ParamSpec
import vllm._C # noqa
import vllm.envs as envs
from vllm.logger import init_logger
from vllm.utils import import_pynvml
from .interface import DeviceCapability, Platform, PlatformEnum, _Backend
@@ -29,13 +29,7 @@ logger = init_logger(__name__)
_P = ParamSpec("_P")
_R = TypeVar("_R")
if pynvml.__file__.endswith("__init__.py"):
logger.warning(
"You are using a deprecated `pynvml` package. Please install"
" `nvidia-ml-py` instead, and make sure to uninstall `pynvml`."
" When both of them are installed, `pynvml` will take precedence"
" and cause errors. See https://pypi.org/project/pynvml "
"for more information.")
pynvml = import_pynvml()
# pytorch 2.5 uses cudnn sdpa by default, which will cause crash on some models
# see https://github.com/huggingface/diffusers/issues/9704 for details