I have already installed on your Linux server cude9.0 server original 10.1 (laboratory), besides pytroch0.3.0 version can't install the other everything is according to the process configuration on this
And on the server NVCC -v results is also NVCC: NVIDIA Cuda (R) compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21: likeness _CDT_2017
Cuda compilation tools, release 9.0, V9.0.176
But enforcement pretrain_detector. Sh will still appear such mistakes:
Traceback (the most recent call last) :
The File "/home/lijiayi/neural - motifs - master/models/train_detector py", line 11, the in & lt; module>
The from lib. Object_detector import ObjectDetector
The File "/home/lijiayi/neural - motifs - master/lib/object_detector py", line 11, the in & lt; module>
The from lib. FPN. NMS. Functions provides. NMS import apply_nms
The File "/home/lijiayi/neural - motifs - master/lib/FPN/NMS/functions provides/NMS. Py", line 4, the in & lt; module>
from .. _ext import NMS
The File "/home/lijiayi/neural - motifs - master/lib/FPN/NMS/_ext/NMS/set py", line 3, the in & lt; module>
From the _nms import lib as _lib, ffi as _ffi
ImportError:/home/lijiayi/neural - motifs - master/lib/FPN/NMS/_ext/NMS/_nms. So: undefined symbol: __cudaRegisterFatBinaryEnd
Excuse me somebody know why
CodePudding user response:
I have just met, I installed the cuda is version 10.1, and Pytorch is 0.4.1 version, both don't match, you can refer to https://www.cnblogs.com/luruiyuan/p/10720581.html, solved no ~ don't know you now