[Installation]: Unmatched torch version between vllm=0.9.2 and vllm-ascend=0.9.2rc1

[GuanlinLee]

Your current environment

pip install vllm==0.9.2 automatically installed the torch=2.7.1.

Then, pip install vllm-ascend==0.9.2rc1 comes with a fatal error that could not find a version that satisfies the requirement torch-npu==2.5.1.post1.dev20250619.

BTW, setting pip config set global.extra-index-url "https://download.pytorch.org/whl/cpu/ https://mirrors.huaweicloud.com/ascend/repos/pypi" will keep the pip looking for index forever long.

How you are installing vllm and vllm-ascend

I followed the steps provided in the documents, as shown in https://vllm-ascend.readthedocs.io/en/latest/installation.html.

[Yikun]

Current installation step, will first install vllm with torch=2.7.1. and then install vllm-ascend==0.9.2rc1 with torch-2.5.1.post1.dev20250619.

There is also a regression test for binary installation:https://github.com/vllm-project/vllm-ascend/blob/main/tests/e2e/doctests/002-pip-binary-installation-test.sh#L36

It's also passed: https://github.com/vllm-project/vllm-ascend/actions/runs/16281460104/job/45971799326#step:5:712

So, I'm not sure what's the specific issue you hitted.

BTW, the torch dependency will be upgraded in #1562 in next release (0.10.0rc1).

[GuanlinLee]

So, I'm not sure what's the specific issue you hitted.

pip install vllm-ascend==0.9.2rc1 comes with a fatal error that could not find a version that satisfies the requirement torch-npu==2.5.1.post1.dev20250619.

I cannot install vllm-ascend with pip install or compile it from binary.

[GuanlinLee]

If I manually install torch-npu with pip install torch-npu==2.5.1, it will uninstall previous torch-2.7.1 and get following errors:

vllm 0.9.2 requires torch==2.7.0; platform_machine == "ppc64le" or platform_machine == "aarch64", but you have torch 2.5.1 which is incompatible.

torchvision 0.22.0 requires torch==2.7.0, but you have torch 2.5.1 which is incompatible.

torchaudio 2.7.0 requires torch==2.7.0, but you have torch 2.5.1 which is incompatible.

llamafactory 0.9.2 requires pandas>=2.0.0, but you have pandas 1.3.5 which is incompatible.

llamafactory 0.9.2 requires tokenizers<=0.21.0,>=0.19.0, but you have tokenizers 0.21.2 which is incompatible.

llamafactory 0.9.2 requires transformers!=4.46.*,!=4.47.*,!=4.48.*,<=4.49.0,>=4.41.2; python_version < "3.10", but you have transformers 4.53.2 which is incompatible.

[Yikun]

could not find a version that satisfies the requirement torch-npu==2.5.1.post1.dev20250619

pip config set global.extra-index-url "https://download.pytorch.org/whl/cpu/ https://mirrors.huaweicloud.com/ascend/repos/pypi"

get following errors

you can ignore the error, because installation is successful

[Yikun]

Any way you can refer the installation progress on : https://github.com/vllm-project/vllm-ascend/actions/runs/16281460104/job/45971799326#step:5:218

[Yikun]

As offline discuss with @GuanlinLee This is because internal network in thier env, so just close this