[Installation]: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv

[li126com]

trafficstars

Your current environment

python 3.10, H100

python -c 'import vllm'
INFO 03-18 16:16:30 __init__.py:183] Automatically detected platform cuda.
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/cpfs01/shared/vllm/vllm/__init__.py", line 6, in <module>
    from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs
  File "/cpfs01/shared/vllm/vllm/engine/arg_utils.py", line 18, in <module>
    from vllm.executor.executor_base import ExecutorBase
  File "/cpfs01/shared/vllm/vllm/executor/executor_base.py", line 13, in <module>
    from vllm.platforms import current_platform
  File "/cpfs01/shared/vllm/vllm/platforms/__init__.py", line 215, in __getattr__
    _current_platform = resolve_obj_by_qualname(
  File "/cpfs01/shared/vllm/vllm/utils.py", line 1894, in resolve_obj_by_qualname
    module = importlib.import_module(module_name)
  File "/cpfs01/shared/env/miniconda/envs/vllm_gpu/lib/python3.10/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "/cpfs01/shared/vllm/vllm/platforms/cuda.py", line 15, in <module>
    import vllm._C  # noqa
ImportError: /cpfs01/shared/vllm/vllm/_C.abi3.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv

How you are installing vllm

conda create -n vllm python=3.10
git clone https://github.com/vllm-project/vllm.git
cd vllm
git checkout v0.7.0
VLLM_USE_PRECOMPILED=1 pip install --editable .

Before submitting a new issue...

• [x] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[shell-nlp]

我在启动时，遇到了同样的问题（I had the same problem when starting up）

[BambiSheng]

Same problem.

[Zhengbang-Yang]

same issue

[2973852520]

Same problem

[TianHongZXY]

same problem

[yantijin]

same here

[Zhengbang-Yang]

My environment can avoid this problem: torch==2.4.0 torchvision==0.19.0 vllm==0.5.5 vllm-flash-attn==2.6.1

[kylesayrs]

I was able to solve this with help from @LucasWilkinson by installing torch==2.6.0

[oncealong]

for my case, VLLM_USE_PRECOMPILED always throw error "vllm/_C.abi3.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv" when run.

VLLM_USE_PRECOMPILED=1 pip install --editable .

VLLM_PRECOMPILED_WHEEL_LOCATION save my time. below command is work for me. VLLM_COMMIT is commit from main branch, which commit you start change python code.

export VLLM_COMMIT=72d9c316d3f6ede485146fe5aabd4e61dbc59069 # use full commit hash from the main branch
export VLLM_PRECOMPILED_WHEEL_LOCATION=https://wheels.vllm.ai/${VLLM_COMMIT}/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
pip install --editable .

[TimerChen]

For torch<2.6, you can try lower version of torchcodec. (ref: https://github.com/pytorch/torchcodec)

pip install torchcodec==0.1

[AndyOO0]

torch==2.6.0

it seems not work

[lyf07]

Has anyone solved this problem?

[aarnphm]

Can you not try with conda? just virtualenv?

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[github-actions[bot]]

This issue has been automatically closed due to inactivity. Please feel free to reopen if you feel it is still relevant. Thank you!