-
Notifications
You must be signed in to change notification settings - Fork 2.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Installation]: Failed to build form source code. Python=3.9 CUDA=12.1 #4201
Comments
Can you try to follow pytorch and build a cuda extension? https://pytorch.org/tutorials/advanced/cpp_extension.html Seems like your environment is broken.
Your python interpreter is here
But it finds torch in some strange path. |
Sorry, I'm not familar with CUDA and don't know how to compile /csrc/moe/topk_softmax_kernels.cu.o. Initially, I thought the strange torch path is caused by not installing torch before building vlllm or the torch is installed with conda not pip. |
Same problem here. BTW, the error when running the build of this wired torch is cause by the incompatible cuda-12.1 and pybind11
|
moe seems to be the cultprint
|
Same here, occurring with the latest commits. Ubuntu 22.04.4, Python 3.10, CUDA 12.2, NVIDIA-SMI 535.171.04, torch-2.2.1, gcc 12.3.0 Just installed torch and cudatoolkit a few days ago with previous vLLM commit which worked. "pip install vllm" works too. Installing from latest commit fails:
|
I am getting the same exact error and I have been able to build previous versions of |
same error, previous version of vllm is fine |
yup same issue here |
the same problem |
I have a similar issue when using gcc-13.2 (manually compiled as Fedora 40 only ships gcc-14) and CUDA 12.4. WHen building vllm, one of the headers in pytorch (
I think, if you adapt the paths, you could use a similar fix to inject patched pybind header files. |
Your current environment
How you are installing vllm
The text was updated successfully, but these errors were encountered: