import error

Question

import error

OCEANOUXIN opened this issue 6 months ago · comments

hi, thanks for your excellent work. When I run nnUNetv2_train 1 3d_fullres all -tr nnUNetTrainerUMambaBot
I don't how to solve it.Could you give me some advise?

Jun · Answer 1 · Thu Mar 28 2024 07:13:47 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,

Could you please test whether mamba is successfully installed?

THESKY · Answer 2 · Fri Mar 29 2024 19:41:21 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,

Could you please test whether mamba is successfully installed?

I encounter this problem too when I import mamba_ssm. It reports
import selective_scan_cuda
ImportError: /data/Tools/anaconda/envs/test/lib/python3.10/site-packages/selective_scan_cuda.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN3c104cuda20CUDACachingAllocator12recordStreamERKNS_7DataPtrENS0_10CUDAStreamE

OCEANOUXIN · Answer 3 · Fri Mar 29 2024 19:56:42 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,

Could you please test whether mamba is successfully installed?

It likes this

THESKY · Answer 4 · Sat Mar 30 2024 10:15:57 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,
Could you please test whether mamba is successfully installed?

It likes this

I think you should create a new env, and follow the orders that authors show. Maybe you install the mamba_ssm first and change the version of torch, but it is wrong. So you might reloaded the mamba_ssm (you might need "pip cache refresh") after the downloading the right torch.Additionally, the version of mamba_ssm might be the 1.1.4

OCEANOUXIN · Answer 5 · Sat Mar 30 2024 19:16:44 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,
Could you please test whether mamba is successfully installed?

It likes this

I think you should create a new env, and follow the orders that authors show. Maybe you install the mamba_ssm first and change the version of torch, but it is wrong. So you might reloaded the mamba_ssm (you might need "pip cache refresh") after the downloading the right torch.Additionally, the version of mamba_ssm might be the 1.1.4

Hi, I really appreciate your reply.
I try to create a new env as the author shows, but I got this:

The author says the version of CUDA 11.7, my version is ：

Does my version have an impact？Isn't the CUDA12.3 backwards compatible?
Best wishes!

THESKY · Answer 6 · Sat Mar 30 2024 20:16:34 GMT+0800 (China Standard Time)

Hi @OCEANOUXIN ,
Could you please test whether mamba is successfully installed?

It likes this

I think you should create a new env, and follow the orders that authors show. Maybe you install the mamba_ssm first and change the version of torch, but it is wrong. So you might reloaded the mamba_ssm (you might need "pip cache refresh") after the downloading the right torch.Additionally, the version of mamba_ssm might be the 1.1.4

Hi, I really appreciate your reply. I try to create a new env as the author shows, but I got this: The author says the version of CUDA 11.7, my version is ： Does my version have an impact？Isn't the CUDA12.3 backwards compatible? Best wishes!

you can first solve the "error :No module name 'packaging'" using pip install packaging.and try it again. If you can't solve this problem, you'd better to download the version of torch author suggested

Jun · Answer 7 · Tue Apr 02 2024 07:23:06 GMT+0800 (China Standard Time)

Hi all,

pip install causal-conv1d>=1.2.0 can solve this issue!