Closed wwq66 closed 4 years ago
My environment: pytorch 0.4.1, cudatoolkit 9.0. After run
python build.py
, when runing train_and_eval.py, I met the mistake.Traceback (most recent call last): File "train_and_eval.py", line 3, in from networks.kd_model import NetModel File "/data/code/structure_knowledge_distillation-master/networks/kd_model.py", line 21, in from networks.pspnet_combine import Res_pspnet, BasicBlock, Bottleneck File "/data/code/structure_knowledge_distillation-master/networks/pspnet_combine.py", line 11, in from libs import InPlaceABN, InPlaceABNSync File "/data/code/structure_knowledge_distillation-master/libs/init.py", line 1, in from .bn import ABN, InPlaceABN, InPlaceABNWrapper, InPlaceABNSync, InPlaceABNSyncWrapper File "/data/code/structure_knowledge_distillation-master/libs/bn.py", line 15, in from .functions import inplace_abn, inplace_abn_sync File "/data/code/structure_knowledge_distillation-master/libs/functions.py", line 5, in from . import _ext File "/data/code/structure_knowledge_distillation-master/libs/_ext/init.py", line 3, in from .__ext import lib as _lib, ffi as _ffi ImportError: /data/code/structure_knowledge_distillation-master/libs/_ext/ext.so: undefined symbol: cudaPopCallConfiguration
Have you worked out any approach to solve the problem?
It is caused by the sync bn. Make sure that the cuda version of your cudatookit and the pytorch is the same. You can refer to this issue. https://github.com/rusty1s/pytorch_scatter/issues/19
Thanks for you reply,I have solved this problem, but unfortunately I have encountered another problem, if it is convenient, I hope you can help me to solve this problem, it would be greatly appreciated. The problem seems to come from /libs/_ext/init.py
At 2019-12-09 09:05:16, "yifan liu" notifications@github.com wrote:
It is caused by the sync bn. Make sure that the cuda version of your cudatookit and the pytorch is the same. You can refer to this issue. rusty1s/pytorch_scatter#19
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.
Are there any attached picture? I can not find your error information.
Is that what you need?
At 2019-12-15 15:47:01, "yifan liu" notifications@github.com wrote:
Are there any attached picture? I can not find your error information.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.
Sorry, I can not see the picture. It looks like the following pictures. Maybe you can send me an email about your question.
yifan.liu04@adelaide.edu.au
My environment: pytorch 0.4.1, cudatoolkit 9.0. After run
python build.py
, when runing train_and_eval.py, I met the mistake.Traceback (most recent call last): File "train_and_eval.py", line 3, in
from networks.kd_model import NetModel
File "/data/code/structure_knowledge_distillation-master/networks/kd_model.py", line 21, in
from networks.pspnet_combine import Res_pspnet, BasicBlock, Bottleneck
File "/data/code/structure_knowledge_distillation-master/networks/pspnet_combine.py", line 11, in
from libs import InPlaceABN, InPlaceABNSync
File "/data/code/structure_knowledge_distillation-master/libs/init.py", line 1, in
from .bn import ABN, InPlaceABN, InPlaceABNWrapper, InPlaceABNSync, InPlaceABNSyncWrapper
File "/data/code/structure_knowledge_distillation-master/libs/bn.py", line 15, in
from .functions import inplace_abn, inplace_abn_sync
File "/data/code/structure_knowledge_distillation-master/libs/functions.py", line 5, in
from . import _ext
File "/data/code/structure_knowledge_distillation-master/libs/_ext/init.py", line 3, in
from .__ext import lib as _lib, ffi as _ffi
ImportError: /data/code/structure_knowledge_distillation-master/libs/_ext/ext.so: undefined symbol: cudaPopCallConfiguration