If you are having a linking error with unresolved "GraphExecKernelNodeSetParams" you can get the fix from here: https://github.com/dkarkoulis/tensorflow-rocm.git
You can merge my pull req https://github.com/rocm-arch/tensorflow-rocm/pull/67
Git Clone URL: | https://aur.archlinux.org/tensorflow-rocm.git (read-only, click to copy) |
---|---|
Package Base: | tensorflow-rocm |
Description: | Library for computation using data flow graphs for scalable machine learning (with ROCM and AVX CPU optimizations) |
Upstream URL: | https://www.tensorflow.org/ |
Licenses: | Apache |
Conflicts: | tensorflow |
Provides: | tensorflow, tensorflow-rocm |
Submitter: | acxz |
Maintainer: | acxz |
Last Packager: | acxz |
Votes: | 11 |
Popularity: | 0.000583 |
First Submitted: | 2020-07-03 20:44 (UTC) |
Last Updated: | 2024-07-21 05:51 (UTC) |
If you are having a linking error with unresolved "GraphExecKernelNodeSetParams" you can get the fix from here: https://github.com/dkarkoulis/tensorflow-rocm.git
You can merge my pull req https://github.com/rocm-arch/tensorflow-rocm/pull/67
Can you place make the (bazel
) download not cluttering the user's home directory (${HOME}/.cache/bazel
) but to be placed unter ${srcdir}/
, e.g. "${srcdir}/.bazel/"
?
Regards and thanks for the package!
Can you please move all the download stuff out from build()
?
If you cannot put it into the source
array, please put it into prepare()
.
Loading:
Fetching repository @pypi; Restarting. 34s
Fetching repository @python_x86_64-unknown-linux-gnu; starting 34s
Fetching https://github.com/indygreg/python-build-standalone/releases/download/20231002/cpython-3.11.6+20231002-x86_64-unknown-linux-gnu-install_only.tar.gz; 15.9 MiB (16,681,357B) 33s
Fetching repository @llvm-raw; starting 65s
Fetching https://storage.googleapis.com/mirror.tensorflow.org/github.com/llvm/llvm-project/archive/49cb1595c1b3ae1de3684fea6148363c15bae12a.tar.gz; 21.8 MiB (22,876,377B) 64s
[...]
Regards and thanks for maintaining!
prepare()
prints out error messages (but continues):
==> Starting prepare()...
bazel 6.1.0
/tmp/makepkg/yay/tensorflow-rocm/PKGBUILD: line 124: /opt/cuda/bin/nvcc: No such file or directory
sed: can't read /usr/include/cudnn_version.h: No such file or directory
does somebody knows about fixes for : error: Illegal instruction detected: Invalid dpp_ctrl value: broadcasts are not supported on GFX10+
@lubosz your compilations is taking ages, and mostly killing the machines I have tried it
Please merge my fixes for ROCm 5.6, llvm 16 and update to tensorflow 2.13: https://github.com/rocm-arch/tensorflow-rocm/pulls
Full working branch: https://github.com/lubosz/tensorflow-rocm/commits/master
Hi. I spent a long time trying to compile tensorflow-rocm but failed. However, you can also run the official rocm/tensorflow docker image, which works for me on Arch with no dependencies other than docker.
Link to the rocm tensorflow docker hub page
In summary:
sudo systemctl start systemd-timesyncd
sudo docker run -it --network=host --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v $HOME/dockerx:/dockerx rocm/tensorflow:latest
My more in depth tutorial here: https://github.com/mpeschel10/test-tensorflow-rocm
i do not use this version of tensorflow anymore, please feel free to adopt.
roctracer is needed to build this package. And after adding roctracer, I get the following error:
ERROR: /build/.cache/bazel/_bazel_builduser/602d407ea92ccabfb83888bc2779e990/external/bazel_tools/platforms/BUILD:84:6: in alias rule @bazel_tools//platforms:linux: Constraints from @bazel_tools//platforms have been removed. Please use constraints from @platforms repository embedded in Bazel, or preferably declare dependency on https://github.com/bazelbuild/platforms. See https://github.com/bazelbuild/bazel/issues/8622 for details.
ERROR: /build/.cache/bazel/_bazel_builduser/602d407ea92ccabfb83888bc2779e990/external/bazel_tools/platforms/BUILD:84:6: Analysis of target '@bazel_tools//platforms:linux' failed
ERROR: /build/tensorflow-amd/src/tensorflow-2.9.2-amd/tensorflow/tools/pip_package/BUILD:277:10: While resolving toolchains for target //tensorflow/tools/pip_package:build_pip_package: Target @local_execution_config_platform//:platform was referenced as a platform, but does not provide PlatformInfo
ERROR: Analysis of target '//tensorflow/tools/pip_package:build_pip_package' failed; build aborted:
INFO: Elapsed time: 30.803s
INFO: 0 processes.
FAILED: Build did NOT complete successfully (7 packages loaded, 8 targets configured)
==> ERROR: A failure occurred in build().
https://github.com/arch4edu/cactus/actions/runs/4182687113/jobs/7246266093
Pinned Comments
acxz commented on 2020-07-03 20:57 (UTC)
Development is on Github: https://github.com/rocm-arch/tensorflow-rocm Please open issues and PRs there instead of commenting.