[BUG]: ModuleNotFoundError: No module named 'colossalai.kernel.op_builder'

[lxing532]

🐛 Describe the bug

I was trying to run: torchrun --standalone --nproc_per_node=2 train_dummy.py --strategy colossalai_zero2 under applications/Chat/examples, and got this error. I tried possible solutions mentioned in other previous relevant bug postings but unfortunately none of them works.

Environment

Python: 3.7.10 Pytorch Version: 1.12.1 Cuda toolkit Version: 11.3

[JThh]

Hi, may I see your colossalai check -i output?

[Ke51n]

Hi, may I see your colossalai check -i output?

I encounter the same error，here is my colossalai check -i output

[Ke51n]

I installed by pip install colossalai ，but when i run the example I encounter a problem “ImportError: cannot import name 'zero_model_wrapper' from 'colossalai.zero'” so I installed by git clone https://github.com/hpcaitech/ColossalAI.git cd ColossalAI pip install .

but encountered the above bug，no module named colossalai.kernel.op_builder

[yifanhunter]

1. Versioning: Make sure pytorch is installed according to the cuda version; 2) Run conda install -c conda-forge gcc=9.5.0 gxx=9.5.0; 3) CUDA_EXT=1 pip install colossalai; 4) ModuleNotFoundError: No module named 'invoke.vendor.decorator', pip install decorator