-
Notifications
You must be signed in to change notification settings - Fork 6.8k
Floating point exception (core dumped) #12105
Comments
We didn't see this issue previously. I think it caused by the environmental issues. Could you 'pip uninstall' all the current mxnet and re-install? If it's possible, please send out the all log for our debugging. |
i try to 'pip uninstall' all the current mxnet and re-install" ,get same result,in fact,i run my docker image in two compute,one of them is run ok,but another one get those err. i don't know how to get other debug info to you, it just come out single message "Floating point exception (core dumped)" |
i run version with 1.0.0, when i get output layer "net_out = self.model.get_outputs()", final come out same error,what should i do
|
Hi @uniquezhengjie could you please provide more context on the system you are running on? Which platform, python version, etc. @szha , could you please take a look in here? @mxnet-label-bot please add [installation, python] here. |
@lanking520 this is my system info, i test Python 3.6.4 and Python 2.7.12 on ubuntu 16.04 and 14.04. i found that when i run pip install mxnet (not mxnet-mkl)or build form source(https://mxnet.incubator.apache.org/install/index.html?platform=Linux&language=Python&processor=CPU) i can import mxnet normally ,but when i inference my model ,it cost mort time(almost 4~ 5 times than mkl in my normal computers(which can use mxnet-mkl)). |
1.0.0 was directly on mklml and not on mkldnn, so the bug likely have already been addressed with mkldnn integration. @uniquezhengjie is there blocker that prevents you from upgrading to the latest version? |
@szha if i run run mxnet-mkl version with 1.0.0, i can import ,but when i get output layer "net_out = self.model.get_outputs()", final come out "Floating point exception". |
Ah I see. Sorry that I misunderstood your question. Does the same happen when you install mxnet without '-mkl' suffix? Also, could you follow the instructions in the issue template and provide the diagnostic information, so that our friends at intel can look at supported instruction sets? Thanks. |
Is the two machine same? |
@pengzhao-intel is different, in fact i test 4 kind cpus |
https://github.com/apache/incubator-mxnet/blob/master/MKLDNN_README.md This is the installation guide :) |
i get error when run "make -j $(nproc) USE_OPENCV=1 USE_MKLDNN=1 USE_BLAS=mkl USE_INTEL_PATH=/opt/intel" cmake /incubator-mxnet/3rdparty/mkldnn -DCMAKE_INSTALL_PREFIX=/incubator-mxnet/3rdparty/mkldnn/install -B/incubator-mxnet/3rdparty/mkldnn/build -DARCH_OPT_FLAGS="-mtune=generic" -DWITH_TEST=OFF -DWITH_EXAMPLE=OFF -- The C compiler identification is GNU 4.9.2 what's problems |
For CUDA versions (as opposed to MKL) this issue seems to be solved in recent daily builds, see: #11911 |
i used command "pip install mxnet-mkl" ,when i import mxnet get Floating point exception (core dumped)
i test mxnet-mkl (1.2.1.post1) mxnet-mkl (1.2.0) get err Floating point exception (core dumped)
mxnet-mkl 1.0.0 import normally
what problem,i run with docker but get same err
The text was updated successfully, but these errors were encountered: