Package Details: caffe2-git 1:0.8.2.r11225.g141d81d09-1

Git Clone URL: https://aur.archlinux.org/caffe2-git.git (read-only)
Package Base: caffe2-git
Description: A new lightweight, modular, and scalable deep learning framework (git version, gpu enabled)
Upstream URL: http://caffe2.ai/
Keywords: ai artificial cuda intelligence nvidia
Licenses: BSD
Conflicts: caffe, caffe-cpu, caffe-cpu-git, caffe-git, caffe2, caffe2-cpu, caffe2-cpu-git
Provides: caffe2
Submitter: dbermond
Maintainer: dbermond
Last Packager: dbermond
Votes: 0
Popularity: 0.000000
First Submitted: 2017-04-29 16:21
Last Updated: 2018-05-11 23:48

Required by (0)

Sources (29)

Latest Comments

dbermond commented on 2018-04-23 17:59

I have added support for building with cuda 9.1. Package is finally building fine again.

Important notes:

  • gcc5 from the AUR is currently needed for building. It does not build with cuda 9.1 + gcc6 (this is a known upstream issue). Be warned that gcc5 from the AUR takes a lot of time to compile.

  • The Caffe2 source code moved to PyTorch repository.

  • Upstream added python3 support on the git master branch. This package now uses python3.

dbermond commented on 2018-01-25 14:27

@barrybingo Thank you for providing some solution, although it sounds rather complicated.

This is a known upstream issue. Currently, caffe2 does not build with CUDA 9. Developers are aware.

Details here: https://github.com/caffe2/caffe2/issues/1459

Unfortunately, package is broken and we should wait for an upstream fix. If someone really wants to use caffe2, it will be needed to downgrade CUDA to version 8.0. Otherwise, use caffe (previous generation).

barrybingo commented on 2018-01-23 13:13

Build failing as per this issue: https://github.com/caffe2/caffe2/issues/1636 See also: https://devtalk.nvidia.com/default/topic/1028112/cuda-setup-and-installation/nvcc-bug-related-to-gcc-6-lt-tuple-gt-header-/

Changing gcc-6/g++-6 to gcc-5/g++-5 for cmake fixed this.<br> Seems a step backwards and build time felt longer.

Build then failed with:

[ 77%] Linking CXX executable ../bin/utility_ops_gpu_test /usr/lib/libtbb.so.2: undefined reference to __cxa_init_primary_exception@CXXABI_1.3.11' /usr/lib/libtbb.so.2: undefined reference tostd::__exception_ptr::exception_ptr::exception_ptr(void)@CXXABI_1.3.11' collect2: error: ld returned 1 exit status make[2]: [caffe2/CMakeFiles/utility_ops_gpu_test.dir/build.make:135: bin/utility_ops_gpu_test] Error 1 make[1]: [CMakeFiles/Makefile2:1248: caffe2/CMakeFiles/utility_ops_gpu_test.dir/all] Error 2 make: ** [Makefile:141: all] Error 2

Fixed this by downgrading package intel-tbb (2018_20171205-1 => 2017_20170412-1) from https://archive.archlinux.org/packages/i/intel-tbb/

Finally working :) gcc-6 (GCC) 6.4.1 20171003 gcc-5 (GCC) 5.5.0

dbermond commented on 2017-09-13 20:50

cuda 8.0.61-3 fixes the glibc 2.26 issue.
This package is now building fine again.

dbermond commented on 2017-09-09 17:09

It seems that CUDA/NVCC 8.0.61 is not compatible with glibc 2.26. As a result, currently this package cannot be compiled with gpu support as it should be.
Use caffe2-cpu-git instead (cpu-only, without gpu support) until there is a repository or upstream fix.
Please leave a reply if you find a workaround.

dbermond commented on 2017-08-10 01:34

Upstream changed the source URL of android-cmake git submodule.
If you have the caffe2 sources on disk and are updating from it, you need to delete the directories 'submodule-android-cmake' and 'src' before making a new build with makepkg.

Otherwise you will get the following error:
error: Server does not allow request for unadvertised object 4c18c932bc18ddbfd55ecf3b1a81bbfb72516ebc
Fetched in submodule path 'third_party/android-cmake', but it did not contain 4c18c932bc18ddbfd55ecf3b1a81bbfb72516ebc. Direct fetching of that commit failed.

dbermond commented on 2017-06-16 00:15

@otaj
Because they have a conflicting file: /usr/include/caffe/proto/caffe.pb.h
If you build caffe and caffe2 you will see that both of them provide this same file.

otaj commented on 2017-06-15 20:26

Hi,

why does this conflicts with original caffe? According to https://github.com/caffe2/caffe2/issues/274 they seem to be going in slightly different directions and definitely in enough different that it is possible to run both on one computer.