search

yaroslavvb / tensorflow-community-wheels

Place to upload links to TensorFlow wheels
453 stars 35 forks source link

TensorFlow 1.13.1, older CPUs with *NO* AVX and *NO* SSE.n. (CUDA 10.0, cuDNN 7.4), Python 3.6, Ubuntu 18.04 (Mint 19.1) #110

Open iridiumblue opened 5 years ago

iridiumblue commented 5 years ago

https://drive.google.com/file/d/1Di3CrEZCEDecYy3lnj1t8hH2RScIFF0u/view?usp=sharing

Thanks to https://github.com/edaustin for helping me with the build.

Abrosimov-a-a commented 5 years ago

Please, share your build command for 1.13. I use: LD_LIBRARY_PATH=/usr/local/cuda/lib64/stubs:${LD_LIBRARY_PATH} tensorflow/tools/ci_build/builds/configured GPU bazel build -c opt --config=cuda --copt=-mavx --copt=-O2 --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" tensorflow/tools/pip_package:build_pip_package But its not worked for me.

iridiumblue commented 5 years ago

I did just did this - bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

I understand bazel can be a little glitchy, though.

I also made two changes to the source, to avoid the dependency on SSE and AVX CPU capabilities -

Removed all checking for SSE and AVX in the file tensorflow/core/platform/cpu_feature_guard.cc.

Did the same for tensorflow/tensorflow.bzl. I don't recall the exact lines, but I can diff the the source if you need me to ...

Did you just try to install the binary wheel I posted? Hopefully that will Just Work for you ...

Abrosimov-a-a commented 5 years ago

It's worked: LD_LIBRARY_PATH=/usr/local/cuda/lib64/stubs:${LD_LIBRARY_PATH} tensorflow/tools/ci_build/builds/configured GPU bazel build -c opt --config=cuda --copt=-march=native --copt=-mtune=native --copt=-O2 tensorflow/tools/pip_package:build_pip_package This command work fine for the same CPU arch on build and runtime machines. For different build and runtime CPU arch you need to replace native. See more. https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html

I can't use your package because i need Python 3.7.

lolouk44 commented 5 years ago

@Abrosimov-a-a If you complete a build with Python 3.7, can you post it? I'd be interested too pls

Abrosimov-a-a commented 5 years ago

My computer has been upgraded. Now I use official packages.

TestPilot23 commented 5 years ago

Could you please build this for TF 1.13.2?

sieumap43 commented 3 years ago

https://drive.google.com/file/d/1Di3CrEZCEDecYy3lnj1t8hH2RScIFF0u/view?usp=sharing

The link doesn't work anymore. Can you give a new link?