Undefined symbol when running test_linear_modules.py

[jhss]

I followed all instructions, but when I ran test_linear_modules.py, the following error happen

ImportError: /home/ubuntu/anaconda3/envs/int/lib/python3.8/site-packages/torch_int/_CUDA.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN3c106detail23torchInternalAssertFailEPKcS2_jS2_RKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE

How can I fix it?

[kyrie2to11]

ImportError: /mnt/afs/user/jarvis/.conda/envs/smoothquant/lib/python3.8/site-packages/torch_int-0.0.0-py3.8-linux-x86_64.egg/torch_int/_CUDA.cpython-38-x86_64-linux-gnu.so:
undefined symbol: cudaOccupancyMaxActiveClusters, version libcudart.so.11.0

similar problem,have you solved it?
if solved, any hint please.
thank you.

[jhss]

In my case, I got above error when I implement custom function in CUTLASS, which might have an error. I think custom function ruined CUDA so file. However, when I switched back to the original CUTLASS repository, the issue was resolved.

I think your problem is different from me, because undefined symbol is cudaOccupancyMaxActiveClusters (In my case unknown garbage value). I'll write down a few things that come to mind right away.

1. check your CUTLASS version is c975e2c
2. check your CUDA version is 11.3

[kyrie2to11]

In my case, I got above error when I implement custom function in CUTLASS, which might have an error. I think custom function ruined CUDA so file. However, when I switched back to the original CUTLASS repository, the issue was resolved.

I think your problem is different from me, because undefined symbol is cudaOccupancyMaxActiveClusters (In my case unknown garbage value). I'll write down a few things that come to mind right away.

1. check your CUTLASS version is c975e2c
2. check your CUDA version is 11.3

Get it. Thanks for your insights!

[Glinttsd]

In my case, I got above error when I implement custom function in CUTLASS, which might have an error. I think custom function ruined CUDA so file. However, when I switched back to the original CUTLASS repository, the issue was resolved.

I think your problem is different from me, because undefined symbol is cudaOccupancyMaxActiveClusters (In my case unknown garbage value). I'll write down a few things that come to mind right away.

1. check your CUTLASS version is c975e2c
2. check your CUDA version is 11.3

Hi, I encountered the same problem as you did. How did you solve it?

I followed the readme and run python tests/test_linear_modules.py without changing anything. The error message is

ImportError: /data/shidi/miniconda3/envs/int/lib/python3.8/site-packages/torch_int-0.0.0-py3.8-linux-x86_64.egg/torch_int/_CUDA.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops10layer_norm4callERKNS_6TensorEN3c108ArrayRefINS5_6SymIntEEERKSt8optionalIS2_ESC_db

[Sakurag1l]

In my case, I got above error when I implement custom function in CUTLASS, which might have an error. I think custom function ruined CUDA so file. However, when I switched back to the original CUTLASS repository, the issue was resolved.
I think your problem is different from me, because undefined symbol is cudaOccupancyMaxActiveClusters (In my case unknown garbage value). I'll write down a few things that come to mind right away.

1. check your CUTLASS version is c975e2c
2. check your CUDA version is 11.3

Get it. Thanks for your insights!

Hi, I have the same problem. Have you solved it?