diff options
author | Marius Bakke <marius@gnu.org> | 2021-08-12 00:30:27 +0200 |
---|---|---|
committer | Marius Bakke <marius@gnu.org> | 2021-08-12 00:30:27 +0200 |
commit | c4133c43c7cfe2476ebfae87f9e4d10d96de9bc7 (patch) | |
tree | 47bd773d2f434384b54e56916c1a287dd8e51511 /gnu/packages/machine-learning.scm | |
parent | ffa01e68859bb7a6daa9fcffdc8d77ca35db4bc0 (diff) | |
parent | 4eb0a5146ae5a195a29c79f586fcc1e58f7fa69b (diff) |
Merge branch 'master' into core-updates-frozen
Conflicts:
gnu/packages/algebra.scm
gnu/packages/games.scm
gnu/packages/golang.scm
gnu/packages/kerberos.scm
gnu/packages/mail.scm
gnu/packages/python.scm
gnu/packages/ruby.scm
gnu/packages/scheme.scm
gnu/packages/tex.scm
gnu/packages/tls.scm
gnu/packages/version-control.scm
Diffstat (limited to 'gnu/packages/machine-learning.scm')
-rw-r--r-- | gnu/packages/machine-learning.scm | 438 |
1 files changed, 412 insertions, 26 deletions
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm index a266c70886..7e7bb89943 100644 --- a/gnu/packages/machine-learning.scm +++ b/gnu/packages/machine-learning.scm @@ -56,6 +56,7 @@ #:use-module (gnu packages check) #:use-module (gnu packages compression) #:use-module (gnu packages cmake) + #:use-module (gnu packages cpp) #:use-module (gnu packages cran) #:use-module (gnu packages databases) #:use-module (gnu packages dejagnu) @@ -64,10 +65,12 @@ #:use-module (gnu packages graphviz) #:use-module (gnu packages gstreamer) #:use-module (gnu packages image) + #:use-module (gnu packages libffi) #:use-module (gnu packages linux) #:use-module (gnu packages llvm) #:use-module (gnu packages maths) #:use-module (gnu packages mpi) + #:use-module (gnu packages ninja) #:use-module (gnu packages ocaml) #:use-module (gnu packages onc-rpc) #:use-module (gnu packages parallel) @@ -86,6 +89,7 @@ #:use-module (gnu packages statistics) #:use-module (gnu packages sqlite) #:use-module (gnu packages swig) + #:use-module (gnu packages tls) #:use-module (gnu packages web) #:use-module (gnu packages xml) #:use-module (gnu packages xorg) @@ -660,6 +664,50 @@ standard data types.") ;; headers, hence the name change. (deprecated-package "python-onnx" onnx)) +(define-public onnx-optimizer + (package + (name "onnx-optimizer") + ;; Note: 0.2.x is *more* recent than 1.5.0. + (version "0.2.6") + (home-page "https://github.com/onnx/optimizer") + (source (origin + (method git-fetch) + (uri (git-reference + (url home-page) + (commit (string-append "v" version)))) + (sha256 + (base32 + "1wkqqdxcxpfbf8zpbdfdd3zz5jkw775g31gyykj11z4y6pp659l6")) + (file-name (git-file-name name version)) + (patches (search-patches "onnx-optimizer-system-library.patch")) + (modules '((guix build utils))) + (snippet '(delete-file-recursively "third_party")))) + (build-system python-build-system) + (arguments (package-arguments onnx)) ;reuse build system tweaks + (native-inputs + `(("cmake" ,cmake) + ("python-pytest" ,python-pytest) + ("python-pytest-runner" ,python-pytest-runner) + ("python-nbval" ,python-nbval) + ("python-coverage" ,python-coverage))) + (inputs + `(("onnx" ,onnx) + ("protobuf" ,protobuf) + ("pybind11" ,pybind11))) + (propagated-inputs + `(("python-numpy" ,python-numpy))) + (synopsis "Library to optimize ONNX models") + (description + "This package provides a C++ and Python library for performing arbitrary +optimizations on ONNX models, as well as a growing list of prepackaged +optimization passes. + +Not all possible optimizations can be directly implemented on ONNX graphs--- +some will need additional backend-specific information---but many can, and the +aim is to provide all such passes along with ONNX so that they can be re-used +with a single function call.") + (license license:expat))) + (define-public rxcpp (package (name "rxcpp") @@ -701,23 +749,22 @@ synchronization, thread-safety, concurrent data structures, and non-blocking I/O.") (license license:asl2.0))) -(define-public gemmlowp-for-tensorflow - ;; The commit hash is taken from "tensorflow/workspace.bzl". - (let ((commit "38ebac7b059e84692f53e5938f97a9943c120d98") - (revision "2")) + +(define-public gemmlowp + (let ((commit "f9959600daa42992baace8a49544a00a743ce1b6") + (version "0.1") + (revision "1")) (package (name "gemmlowp") - (version (git-version "0" revision commit)) + (version (git-version version revision commit)) + (home-page "https://github.com/google/gemmlowp") (source (origin - (method url-fetch) - (uri (string-append "https://mirror.bazel.build/" - "github.com/google/gemmlowp/archive/" - commit ".zip")) - (file-name (string-append "gemmlowp-" version ".zip")) + (method git-fetch) + (uri (git-reference (url home-page) (commit commit))) + (file-name (git-file-name name version)) (sha256 (base32 - "0n56s2g8hrssm4w8qj1v58gfm56a04n9v992ixkmvk6zjiralzxq")))) - (build-system cmake-build-system) + "1hzfhlhzcb827aza6a7drydc67dw5fm3qfqilb9ibskan8dsf0c6")))) (arguments `(#:configure-flags (list ,@(match (%current-system) @@ -737,18 +784,15 @@ I/O.") (inc (string-append out "/include/"))) (install-file "../build/libeight_bit_int_gemm.so" lib) (for-each (lambda (dir) - (let ((target (string-append inc "/" dir))) - (mkdir-p target) + (let ((target + (string-append inc "/gemmlowp/" dir))) (for-each (lambda (h) (install-file h target)) (find-files (string-append "../" dir) "\\.h$")))) '("meta" "profiling" "public" "fixedpoint" - "eight_bit_int_gemm" "internal")) - #t)))))) - (native-inputs - `(("unzip" ,unzip))) - (home-page "https://github.com/google/gemmlowp") + "eight_bit_int_gemm" "internal")))))))) + (build-system cmake-build-system) (synopsis "Small self-contained low-precision GEMM library") (description "This is a small self-contained low-precision @dfn{general matrix @@ -758,6 +802,46 @@ at most 8 bits. To avoid overflow, results are internally accumulated on more than 8 bits, and at the end only some significant 8 bits are kept.") (license license:asl2.0)))) +(define-public gemmlowp-for-tensorflow + ;; The commit hash is taken from "tensorflow/workspace.bzl". + (let ((commit "38ebac7b059e84692f53e5938f97a9943c120d98") + (revision "2")) + (package + (inherit gemmlowp) + (version (git-version "0" revision commit)) + (source (origin + (method url-fetch) + (uri (string-append "https://mirror.bazel.build/" + "github.com/google/gemmlowp/archive/" + commit ".zip")) + (file-name (string-append "gemmlowp-" version ".zip")) + (sha256 + (base32 + "0n56s2g8hrssm4w8qj1v58gfm56a04n9v992ixkmvk6zjiralzxq")))) + (arguments + (substitute-keyword-arguments (package-arguments gemmlowp) + ((#:phases phases) + `(modify-phases ,phases + (replace 'install + (lambda* (#:key outputs #:allow-other-keys) + (let* ((out (assoc-ref outputs "out")) + (lib (string-append out "/lib/")) + (inc (string-append out "/include/"))) + (install-file "../build/libeight_bit_int_gemm.so" lib) + (for-each (lambda (dir) + ;; Note: Install headers straight into + ;; $includedir instead of $includedir/gemmlowp. + (let ((target (string-append inc "/" dir))) + (for-each (lambda (h) + (install-file h target)) + (find-files (string-append "../" dir) + "\\.h$")))) + '("meta" "profiling" "public" "fixedpoint" + "eight_bit_int_gemm" "internal"))))))))) + (native-inputs + `(("unzip" ,unzip))) + (properties '((hidden? . #t)))))) + (define-public dlib (package (name "dlib") @@ -2030,6 +2114,175 @@ together building blocks and a subclassing API with an imperative style for advanced research.") (license license:asl2.0))) +(define-public tensorflow-lite + (package + (name "tensorflow-lite") + (version "2.5.0") + (source + (origin + (method git-fetch) + (uri (git-reference + (url "https://github.com/tensorflow/tensorflow") + (commit (string-append "v" version)))) + (file-name (git-file-name name version)) + (sha256 + (base32 + "1jdw2i1rq06zqd6aabh7bbm0avsg4pygnfmd7gviv0blhih9054l")))) + (build-system cmake-build-system) + (arguments + `(#:tests? #false ; no "check" target + #:build-type "Release" + #:configure-flags + (list + "-DTFLITE_ENABLE_GPU=OFF" + "-DTFLITE_ENABLE_RUY=OFF" + + ;; TODO: The build system attempts to build xnnpack from source. We + ;; would like to use our xnnpack package here, but this requires more + ;; work. + "-DTFLITE_ENABLE_XNNPACK=OFF" + + ;; Pretend we've already fetched abseil. We won't actually build it + ;; but use the existing package. + "-Dabseil-cpp_POPULATED=TRUE" + + ;; Don't fetch the sources. We have already built flatbuffers. + "-Dflatbuffers_POPULATED=TRUE" + + "-DFFT2D_SOURCE_DIR=/tmp/fft2d" + "-Dneon2sse_SOURCE_DIR=/tmp/neon2sse" + "-Dneon2sse_BINARY_DIR=/tmp/neon2sse-bin" + "-DFARMHASH_SOURCE_DIR=/tmp/farmhash" + "-Dgemmlowp_SOURCE_DIR=/tmp/gemmlowp" + (string-append "-DRUY_SOURCE_DIR=" + (assoc-ref %build-inputs "ruy-src"))) + #:phases + (modify-phases %standard-phases + (add-after 'unpack 'chdir + (lambda _ (chdir "tensorflow/lite"))) + (add-after 'chdir 'copy-sources + (lambda* (#:key inputs #:allow-other-keys) + ;; Use external cmake finders instead of these stubs that won't + ;; find anything but the bundled sources. + (delete-file "tools/cmake/modules/Findabsl.cmake") + (delete-file "tools/cmake/modules/Findeigen.cmake") + + (substitute* "CMakeLists.txt" + (("find_package\\(eigen REQUIRED") + "find_package(eigen REQUIRED NAMES Eigen3")) + (substitute* "tools/cmake/modules/Findflatbuffers.cmake" + (("get_target_property.*") + (format #false "set(FLATBUFFERS_INCLUDE_DIRS ~a/include)\n" + (assoc-ref inputs "flatbuffers")))) + + ;; Don't fetch source code; we already have everything we need. + (substitute* '("tools/cmake/modules/fft2d.cmake" + "tools/cmake/modules/ruy.cmake" + "tools/cmake/modules/farmhash.cmake" + "tools/cmake/modules/neon2sse.cmake" + "tools/cmake/modules/gemmlowp.cmake") + (("OverridableFetchContent_Populate.*") "")) + + (mkdir-p "/tmp/farmhash") + (with-directory-excursion "/tmp/farmhash" + (invoke "tar" "--strip-components=1" + "-xf" (assoc-ref inputs "farmhash-src"))) + + (mkdir-p "/tmp/fft2d") + (with-directory-excursion "/tmp/fft2d" + (invoke "tar" "--strip-components=1" + "-xf" (assoc-ref inputs "fft2d-src"))) + + (copy-recursively (assoc-ref inputs "neon2sse-src") + "/tmp/neon2sse/") + (copy-recursively (assoc-ref inputs "gemmlowp-src") + "/tmp/gemmlowp/"))) + (add-after 'copy-sources 'prepare-shared-library-build + (lambda _ (chdir "c"))) + (replace 'install + (lambda* (#:key outputs #:allow-other-keys) + (let* ((out (assoc-ref outputs "out")) + (lib (string-append out "/lib")) + (headers (string-append out "/include/tensorflow/lite"))) + (install-file "../build/libtensorflowlite_c.so" lib) + (with-directory-excursion ".." + (for-each + (lambda (file) + (let ((target-dir (string-append headers "/" (dirname file)))) + (install-file file target-dir))) + (find-files "." "\\.h$"))))))))) + (inputs + `(("abseil-cpp" ,abseil-cpp) + ("eigen" ,eigen-for-tensorflow-lite) + ("flatbuffers" ,flatbuffers) + ("python" ,python))) + (native-inputs + `(("pkg-config" ,pkg-config) + ("gemmlowp-src" + ;; The commit hash is taken from + ;; "tensorflow/lite/tools/cmake/modules/gemmlowp.cmake". + ,(let ((commit "fda83bdc38b118cc6b56753bd540caa49e570745")) + (origin + (method git-fetch) + (uri (git-reference + (url "https://github.com/google/gemmlowp") + (commit commit))) + (file-name (git-file-name "gemmlowp" (string-take commit 8))) + (sha256 + (base32 + "1sbp8kmr2azwlvfbzryy1frxi99jhsh1nc93bdbxdf8zdgpv0kxl"))))) + ("neon2sse-src" + ,(let ((commit "a1652fd5253afbf3e39357b012974f93511f6108")) + (origin + (method git-fetch) + (uri (git-reference + (url "https://github.com/intel/ARM_NEON_2_x86_SSE") + (commit commit))) + (file-name (git-file-name "neon2sse" (string-take commit 8))) + (sha256 + (base32 + "1q8gkxag9wlnwdwad2pclsrkwzrdjy94hyrkayrsvxyj7szb5y8i"))))) + ("farmhash-src" + ,(let ((commit "816a4ae622e964763ca0862d9dbd19324a1eaf45")) + (origin + (method url-fetch) + (uri (string-append + "https://mirror.bazel.build/github.com/google/farmhash/archive/" + commit ".tar.gz")) + (file-name (git-file-name "farmhash" (string-take commit 8))) + (sha256 + (base32 + "185b2xdxl4d4cnsnv6abg8s22gxvx8673jq2yaq85bz4cdy58q35"))))) + ("fft2d-src" + ,(origin + (method url-fetch) + (uri (string-append "https://storage.googleapis.com/" + "mirror.tensorflow.org/" + "www.kurims.kyoto-u.ac.jp/~ooura/fft2d.tgz")) + (file-name "fft2d.tar.gz") + (sha256 + (base32 + "1jfflzi74fag9z4qmgwvp90aif4dpbr1657izmxlgvf4hy8fk9xd")))) + ("ruy-src" + ,(let ((commit "9c56af3fce210a8a103eda19bd6f47c08a9e3d90")) + (origin + (method git-fetch) + (uri (git-reference + (url "https://github.com/google/ruy") + (commit commit) + (recursive? #true))) + (file-name (git-file-name "ruy" (string-take commit 8))) + (sha256 + (base32 + "1cfd5gk6kaj8kbl3h98gx1ap8czd59y6p8qq8nr28fklpyzf5cis"))))))) + (home-page "https://tensorflow.org") + (synopsis "Machine learning framework") + (description + "TensorFlow is a flexible platform for building and training machine +learning models. This package provides the \"lite\" variant for mobile +devices.") + (license license:asl2.0))) + (define-public python-iml (package (name "python-iml") @@ -2194,8 +2447,8 @@ that: (define-public gloo (let ((version "0.0.0") ; no proper version tag - (commit "ca528e32fea9ca8f2b16053cff17160290fc84ce") - (revision "0")) + (commit "c22a5cfba94edf8ea4f53a174d38aa0c629d070f") + (revision "1")) (package (name "gloo") (version (git-version version revision commit)) @@ -2208,18 +2461,20 @@ that: (file-name (git-file-name name version)) (sha256 (base32 - "1q9f80zy75f6njrzrqkmhc0g3qxs4gskr7ns2jdqanxa2ww7a99w")))) + "1crmqgybzkgkpbmcx16912gsl5qsj49swa0ikx6mhqgph0chrh11")))) (build-system cmake-build-system) (native-inputs `(("googletest" ,googletest))) + (inputs + `(("openssl" ,openssl))) (arguments `(#:configure-flags '("-DBUILD_TEST=1") #:phases (modify-phases %standard-phases (replace 'check - (lambda _ - (invoke "make" "gloo_test") - #t))))) + (lambda* (#:key tests? #:allow-other-keys) + (when tests? + (invoke "make" "gloo_test"))))))) (synopsis "Collective communications library") (description "Gloo is a collective communications library. It comes with a @@ -2265,7 +2520,7 @@ general non-linear dimension reduction.") (package (name "xnnpack") (version (git-version version revision commit)) - (home-page "https://github.com/google/XNNPACK") + (home-page "https://github.com/google/XNNPACK") ;fork of QNNPACK (source (origin (method git-fetch) (uri (git-reference (url home-page) (commit commit))) @@ -2305,3 +2560,134 @@ high-level machine learning frameworks, such as TensorFlow Lite, TensorFlow.js, PyTorch, and MediaPipe.") (license license:bsd-3)))) +(define-public python-pytorch + (package + (name "python-pytorch") + (version "1.9.0") + (source (origin + (method git-fetch) + (uri (git-reference + (url "https://github.com/pytorch/pytorch") + (commit (string-append "v" version)) + (recursive? #t))) + (file-name (git-file-name name version)) + (sha256 + (base32 + "0cznsh68hwk5761gv7iijb4g6jgjpvs3bbixwpzzmkbkbn2q96c1")) + (patches (search-patches "python-pytorch-system-libraries.patch" + "python-pytorch-runpath.patch")) + (modules '((guix build utils))) + (snippet + '(begin + ;; XXX: Let's be clear: this package is a bundling fest. We + ;; delete as much as we can, but there's still a lot left. + (for-each (lambda (directory) + (delete-file-recursively + (string-append "third_party/" directory))) + '("benchmark" "cpuinfo" "eigen" + + ;; FIXME: QNNPACK (of which XNNPACK is a fork) + ;; needs these. + ;; "FP16" "FXdiv" "gemmlowp" "psimd" + + "gloo" "googletest" "ios-cmake" + "onnx" "protobuf" "pthreadpool" + "pybind11" "python-enum" "python-peachpy" + "python-six" "tbb" "XNNPACK" "zstd")) + + ;; Adjust references to the onnx-optimizer headers. + (substitute* "caffe2/onnx/backend.cc" + (("onnx/optimizer/") + "onnxoptimizer/")))))) + (build-system python-build-system) + (arguments + '(#:phases (modify-phases %standard-phases + (add-before 'build 'use-system-libraries + (lambda* (#:key outputs #:allow-other-keys) + ;; Tell 'setup.py' to let 'CMakeLists.txt' know that we + ;; want to use "system libraries" instead of the bundled + ;; ones. + (setenv "USE_SYSTEM_LIBS" "1") + + ;; XXX: Disable that for simplicity for now. + (setenv "USE_FBGEMM" "0"))) + (add-before 'build 'make-things-writable + (lambda _ + ;; The 'build_caffe2' function in + ;; 'tools/build_pytorch_libs.py', called from the + ;; top-level 'setup.py', needs write access to this + ;; directory. + (for-each make-file-writable + (find-files "caffe2/proto" "." + #:directories? #t)))) + (replace 'check + (lambda* (#:key inputs outputs tests? #:allow-other-keys) + ;; Run the test suite following the instructions in + ;; 'CONTRIBUTING.md'. XXX: Unfortunately this doesn't + ;; work, unless you set PYTHONPATH presumably. + (when tests? + (let ((python-site (site-packages inputs outputs))) + (setenv "PYTHONPATH" + (string-append python-site ":" + (getenv "PYTHONPATH"))) + (invoke "python" "test/run_test.py"))))) + (add-after 'install 'remove-test-executables + (lambda* (#:key inputs outputs #:allow-other-keys) + ;; Remove test executables, but keep other executables + ;; such as 'torch_shm_manager' and and .so files such as + ;; 'libtorch_global_deps.so'. + (let ((python-site (site-packages inputs outputs))) + (for-each delete-file + (find-files python-site + "(^test_cpp_rpc|_test)$")))))) + + ;; XXX: Tests attempt to download data such as + ;; <https://raw.githubusercontent.com/pytorch/test-infra/master/stats/slow-tests.json>. + #:tests? #f)) + (native-inputs + `(("cmake" ,cmake) + ("ninja" ,ninja))) + (inputs + `(("eigen" ,eigen) + ;; ("fmt" ,fmt) + ("fp16" ,fp16) + ("gemmlowp" ,gemmlowp) + ("googletest" ,googletest) + ("googlebenchmark" ,googlebenchmark) + ("gloo" ,gloo) + ("openblas" ,openblas) + ("openmpi" ,openmpi) + ("pthreadpool" ,pthreadpool) + ("protobuf" ,protobuf) + ("pybind11" ,pybind11) + ("sleef" ,sleef) + ("xnnpack" ,xnnpack) + ("zstd" ,zstd))) + (propagated-inputs + `(("python-astunparse" ,python-astunparse) + ("python-numpy" ,python-numpy) + ("python-pyyaml" ,python-pyyaml) + ("python-cffi" ,python-cffi) + ("python-peachpy" ,python-peachpy) + ("python-typing-extensions" ,python-typing-extensions) + ("python-future" ,python-future) + ("python-six" ,python-six) + ("python-requests" ,python-requests) + ("onnx" ,onnx) ;propagated for its Python modules + ("onnx-optimizer" ,onnx-optimizer) + ("cpuinfo" ,cpuinfo))) + (home-page "https://pytorch.org/") + (synopsis "Python library for tensor computation and deep neural networks") + (description + "PyTorch is a Python package that provides two high-level features: + +@itemize +@item tensor computation (like NumPy) with strong GPU acceleration; +@item deep neural networks (DNNs) built on a tape-based autograd system. +@end itemize + +You can reuse Python packages such as NumPy, SciPy, and Cython to extend +PyTorch when needed. + +Note: currently this package does not provide GPU support.") + (license license:bsd-3))) |