::~# pytest tests/
======================= test session starts ==============================
platform linux -- Python 3.9.12, pytest-7.1.1, pluggy-1.0.0
rootdir: /root/vllm
plugins: anyio-3.5.0, forked-1.6.0, asyncio-0.23.5
asyncio: mode=strict
collected 0 items / 1 error
============================= ERRORS==============================
___________________ ERROR collecting test session _______________________
../anaconda3/lib/python3.9/importlib/**init**.py:127: in import_module
return _bootstrap._gcd_import(name[level:], package, level)
:1030: in _gcd_import
???
:1007: in _find_and_load
???
:986: in _find_and_load_unlocked
???
:680: in _load_unlocked
???
../anaconda3/lib/python3.9/site-packages/_pytest/assertion/rewrite.py:168: in exec_module
exec(co, module.**dict**)
tests/lora/conftest.py:15: in
from vllm.model_executor.layers.sampler import Sampler
vllm/model_executor/**init**.py:2: in
from vllm.model_executor.model_loader import get_model
vllm/model_executor/model_loader.py:10: in
from vllm.model_executor.weight_utils import (get_quant_config,
vllm/model_executor/weight_utils.py:18: in
from vllm.model_executor.layers.quantization import (get_quantization_config,
vllm/model_executor/layers/quantization/**init**.py:4: in
from vllm.model_executor.layers.quantization.awq import AWQConfig
vllm/model_executor/layers/quantization/awq.py:6: in
from vllm._C import ops
E ImportError: /root/vllm/vllm/_C.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZNSt15__exception_ptr13exception_ptr9_M_addrefEv
===================== short test summary info=====================
ERROR - ImportError: /root/vllm/vllm/_C.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZNSt15__exception_pt...
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: 1 error during collection !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
=======================1 error in 0.51s=============================
There is an identical error when running offline_inference.py and the solution I saw is to change gcc from 11 to 10.4, I tried 10.5, it does not work. I wonder if there is a more universal way to solve this issue.
1条答案
按热度按时间tcbh2hod1#
当我尝试从主分支使用
pip install -e .
安装时,遇到了相同的错误,使用的是torch==2.1.2和cuda12.1。