Hi, I’ve installed Pytorch several times, but no luck this time. I need to install the up-to-date codes, so I clone the source and installed every pre-condition software. But when I python setup.py install I just found such error. Any idea?
running install
running build_deps
cmake: /lib64/libstdc++.so.6: version GLIBCXX_3.4.20' not found (required by cmake) cmake: /lib64/libstdc++.so.6: version CXXABI_1.3.9’ not found (required by cmake)
cmake: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21’ not found (required by cmake)
OS: CentOS, I use gcc 4.8 then update to 4.9 + python 3.6.2 + Cuda 8.0 + cudnn 9.0. My GPU is 1080 Ti.
Thanks in advanced.
For the record, if you (or somebody else) encounters other strange build errors (particularly stuff about at_exit, or pthread not found), then you might want to try this:
Instead of installing cmake and gcc via conda, install them via your system. I’m doing this on ubuntu 16.04.
So my setup looks like:
sudo apt-get install g++ cmake
conda install numpy pyyaml mkl setuptools cffi # notice g++ and cmake removed from here
This was mentioned, I think by Sean Naren on a warp-ctc build issue, but thought I might mention it here too.
gcc-6 cannot compile source code correctly in my machine.
Linking C shared library libTH.so
/usr/bin/ld: /home/xxx/anaconda3/lib/gcc/x86_64-pc-linux-gnu/6.1.0/crtbeginS.o: unrecognized relocation (0x2a) in section `.text’
/usr/bin/ld: final link failed: Bad value