* gnu/packages/machine-learning.scm (python-imbalanced-learn): Update to
0.11.0.
[arguments]<#:test-flags>: Move broken tests from "unbreak-tests" phase to
here.
<#:phases>: Add back some deleted python-keras and tensorflow tests.
[native-inputs]: Add python-keras, python-numpydoc and tensorflow.
Change-Id: Ibc2827e27ee23e0ffb491cb791bba4d1a3f1109d
Signed-off-by: Arun Isaac <arunisaac@systemreboot.net>
* gnu/packages/algebra.scm (eigen-for-tensorflow-lite): Remove variable.
* gnu/packages/machine-learning.scm (tensorflow-lite): Update to 2.12.1
[arguments]<configure-flags>: Turn on GPU and RUY use; Use cmake to
find packages already in Guix as dependencies for absl-cpp, eigen,
flatbuffer, neon2ssl, cpuinfo, ruy.
[arguments]<phases>: Remove operations setting up absl-cpp, eigen, ruy,
and neon2ssl as native inouts as now we use Guix's packages of these
as dependencies. Do default build phase.
Add steps to install C shared library and benchmark_model tool.
[inputs]: Add cpuinfo, eigen, fp16, mesa-header, opencl,
pthreadpool, ruy, vulkan and xnnpack as explicit inputs.
[native-inputs]: Remove local setup of neon2ssl and ruy
as we now use Guix's packages of these as explicit dependencies.
* gnu/packages/serialization.scm (flatbuffers-next-shared): New variable,
flatbuffers-next built with -fPIC as needed by tensorflow-lite.
Signed-off-by: Ricardo Wurmus <rekado@elephly.net>
'CMake Error at src/cpu/rv64/CMakeLists.txt:36 (message):
Only sequential runtime is now supported for a RISC-V CPU'.
* gnu/packages/machine-learning.scm (oneapi-dnnl)[arguments]: Adjust
configure-flags when building for riscv64-linux to fix build.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
* gnu/packages/machine-learning.scm (tensorflow)[arguments]: Add one more
patch of "collections" to "collections.abc" to the 'python3.10-compatibility
phase.
Subversion keyword expansion is potentially non-reproducible as some of
them expand time strings relative to the local time zone:
https://issues.guix.gnu.org/43442#18
In practice this is not a problem in Guix since Subversion checkouts
happen in an isolated environment using the "default timezone" (UTC).
However, Software Heritage disables keyword expansion for this very
reason. By following suit, we make sure content can be retrieved from
there.
* guix/build/svn.scm (svn-fecth): Pass "--ignore-keywords" to
Subversion.
* guix/build-system/texlive.scm (%texlive-date): New variable.
* gnu/packages/java.scm (java-geronimo-xbean-reflect)
(java-geronimo-genesis-2.1): Update the source hash.
* gnu/packages/machine-learning.scm (ghmm): Likewise.
* gnu/packages/video.scm (libsmpeg, libsmpeg-with-sdl1): Likewise.
* gnu/packages/tex.scm (texlive-bin): Update the hash of the
"texlive-scripts" input, and a add a new phase that imitates
Subversion keyword expansion for scripts that need it.
(texlive-latex-base): Update the hash of the "texlive-luatexconfig"
native input.
(texlive-hyphen-base, texlive-dvipdfmx, texlive-dvips, texlive-cm)
(texlive-tex-plain, texlive-kpathsea, texlive-latexconfig)
(texlive-tetex, texlive-pdftex, texlive-xetex): Update the source
hash.
Co-authored-by: Ludovic Courtès <ludo@gnu.org>
* gnu/packages/machine-learning.scm (kaldi-for-vosk)[inputs]: Remove spurious
semicolon.
[arguments]: Use LIST instead of double quoting; replace LET* with LET; adjust
indentation.
The variant of the fastapi package must match that of
python-pytorch-lightning.
* gnu/packages/machine-learning.scm (python-lightning-cloud)
[propagated-inputs]: Replace python-fastapi with
python-fastapi-for-pytorch-lightning.
Timothee Mathieu