[Bug] Fail to build MLC_LLM from source

Louym commented 2 months ago

🐛 Bug

Fail to build MLC_LLM from source

To Reproduce

Steps to reproduce the behavior:

Just follow the steps in the documentation to build MLC_LLM from source and the error arises when running cmake .. && cmake --build . --parallel $(nproc) && cd .. I tried to build with Flash Attention and without, all leading to the same result.

Expected behavior

Build failed!

Environment

Platform (e.g. WebGPU/Vulkan/IOS/Android/CUDA): CUDA
Operating system (e.g. Ubuntu/Windows/MacOS/...): Ubuntu-22.04
Device (e.g. iPhone 12 Pro, PC+RTX 3090, ...): RTX 4090&RTX 3090
How you installed MLC-LLM (conda, source): source
How you installed TVM-Unity (pip, source): source
Python version (e.g. 3.10): 3.11.10
GPU driver version (if applicable): 535.161.07
CUDA/cuDNN version (if applicable): 11.8
TVM Unity Hash Tag (python -c "import tvm; print('\n'.join(f'{k}: {v}' for k, v in tvm.support.libinfo().items()))", applicable if you compile models):
Any other relevant information:

Additional context

I wonder how to solve the problem and if I can solve it without changing the version of cuda and so on.

Louym commented 2 months ago

@ArthurMLi @AhmadMahmoodRana Thank you for your quick reply. However, it seems I already have gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 and clang 14.0.0-1ubuntu1.1.

Louym commented 2 months ago

And the website https://bit.ly/4gwC82r seems to be blocked?

ArthurMLi commented 2 months ago

And the website https://bit.ly/4gwC82r seems to be blocked? Louym pls don't download this link, i got hacked and the hacker send this. If you downloaded pls format your pc.

Louym commented 2 months ago

And the website https://bit.ly/4gwC82r seems to be blocked? Louym pls don't download this link, i got hacked and the hacker send this. If you downloaded pls format your pc.

Thank goodness I didn’t download it！

MasterJH5574 commented 2 months ago

@Louym Thanks for reporting the build failure. From the screenshot we can tell little as it does not show the details of the error. If you could share more about the log, that will be informative and helpful. Thanks!

Louym commented 2 months ago

@MasterJH5574 Thank you for your reply. There is no error in the terminal, just some warnings saying that /mlc-llm/3rdparty/tvm/3rdparty/flashinfer/include/flashinfer/attention/prefill.cuh:1979:47: warning: comparison of integer expressions of different signedness: ‘int’ and ‘long unsigned int’ [-Wsign-compare] And there is no errorlog or something alike. Even building with verbose does not generate more useful information. To be honest, I am not very familiar with building and can you tell me where I can find more information?

Louym commented 2 months ago

Has anyone else encountered these issues? I’d appreciate any solutions!

MasterJH5574 commented 2 months ago

Hi @Louym sorry for the delayed response. I'm a bit surprised that there is not error in the log. Do you mind sharing the full log? I'm not 100% sure but the CUDA 11.8 you are using can be a reason, as we have deprecated the 11.8 support, given dependency on some features in CUDA 12.

Louym commented 2 months ago

@MasterJH5574 Thank you so much for your reply! There seems no CMakeError.log or CMakeOutput.log in the CMakeFiles. Cmake itself fails possibly. Anyway, I listen to your advice and successfully run MLC with cuda 12!

mlc-ai / mlc-llm