rcca compile error - Githubissues

Serge-weihao / CCNet-Pure-Pytorch

Criss-Cross Attention (2d&3d) for Semantic Segmentation in pure Pytorch with a faster and more precise implementation.

MIT License

183 stars 21 forks source link

rcca compile error #7

Open Alva-2020 opened 3 years ago

Alva-2020 commented 3 years ago

File "/home/zhangli/anaconda3/envs/py36/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 1538, in _run_ninja_build raise RuntimeError(message) from e RuntimeError: Error building extension 'rcca'

Serge-weihao commented 3 years ago

are you using torch 1.0+?

Alva-2020 commented 3 years ago

are you using torch 1.0+? Yes.

Alva-2020 commented 3 years ago

Could you please give me some suggestions?

Serge-weihao commented 3 years ago

@Alva-2020 you can try python lower than 3.6 and cuda 10.0

Serge-weihao commented 3 years ago

python 3.6 with cuda10.0 can work

johndpope commented 3 years ago

fyi

I saw the original repo deleted Delete rcca.cpython-36m-x86_64-linux-gnu.so 4c7923b

https://github.com/speedinghzl/CCNet/tree/pytorch-1.1

UPDATE - actually same error.

https://github.com/speedinghzl/CCNet/issues/98

UPDATE installing apex withou cuda / also didn't help pip install -v --disable-pip-version-check --no-cache-dir ./

Serge-weihao commented 3 years ago

@johndpope .so file is generated by compiling the cpp cuda extension, I tried that more than one year ago using lower version python and pytorch-1.1 cuda10.0. if you only use from .CC import CC_module as CrissCrossAttention, you need not to compile the cuda extension

8secz-johndpope commented 3 years ago

I can't use cuda 10 - as I have a later graphics card - 3090 :( will see if I can get around it.