Skip to Content

libtorch v2.12.0

Created
Updated

Tensors and Dynamic neural networks in Python with strong GPU acceleration

# License

Unspecified

# Supported Platforms

(windows & !static) | osx | linux

  • linux
  • osx
  • windows-dynamic
  • uwp
  • windows-static

# Features

Default feature: blas

# blas default

BLAS/LAPACK backend (openblas on linux/windows, Accelerate on macOS). On by default; disable with [core,...] when using [mkl] instead.

Dependencies:
and 4 transitive dependencies:
Host Dependencies:
2 transitive dependencies:

# cuda

Build with CUDA GPU backend

Supports: (windows & x64 & !staticcrt) | (linux & x64)
  • linux
  • windows-x64-dynamic
  • osx
  • windows-arm64-dynamic
  • windows-x86-dynamic
Dependencies:
and 9 transitive dependencies:
Host Dependencies:
2 transitive dependencies:

# dist

Use distributed training/inference (Gloo, MPI, libuv, TensorPipe)

Supports: (linux & x64) | windows
  • linux
  • windows-dynamic
  • osx
Dependencies:
and 4 transitive dependencies:
Host Dependencies:
6 transitive dependencies:

# gflags

Build with gflags

Dependencies:

No transitive dependencies.

Host Dependencies:
2 transitive dependencies:

# glog

Build with glog

Dependencies:
and one transitive dependency:
Host Dependencies:
2 transitive dependencies:

# llvm

Build with LLVM

Dependencies:
and 3 transitive dependencies:
Host Dependencies:
3 transitive dependencies:

# mkl

Intel performance backend: use oneMKL for BLAS/LAPACK and enable oneDNN (MKLDNN) CPU acceleration via ideep. Mutually exclusive with [blas] (the openblas default-feature) — install as libtorch[core,mkl,...] to skip the openblas backend, otherwise both backends get built and only MKL is linked.

Supports: (windows | linux) & x64 & !uwp
  • linux
  • windows-x64-dynamic
  • osx
  • windows-arm64-dynamic
  • windows-x86-dynamic
Dependencies:

No transitive dependencies.

Host Dependencies:
2 transitive dependencies:

# vulkan

Build with Vulkan GPU backend

Dependencies:
and one transitive dependency:
Host Dependencies:
and 5 transitive dependencies:

# Dependencies

and 18 transitive dependencies:

# Host Dependencies

# Dependents

No dependents.

# Host Dependents

No dependents.

# Contributors

  • PARK DongHa's avatar PARK DongHa
  • jim wang's avatar jim wang
  • Lily Wang's avatar Lily Wang
  • MonicaLiu's avatar MonicaLiu
  • Ómar Högni Guðmarsson's avatar Ómar Högni Guðmarsson
  • Alexander Neumann's avatar Alexander Neumann
  • Jia Yue Hua's avatar Jia Yue Hua
  • Billy O'Neal's avatar Billy O'Neal
  • miyanyan's avatar miyanyan
  • JoergAtGithub's avatar JoergAtGithub
  • Stefano Sinigardi's avatar Stefano Sinigardi
  • Mengna-Li's avatar Mengna-Li
  • Steve Brain's avatar Steve Brain

# Changelog

  • 32305df [libtorch] update to 2.12 with dependencies (#52108)
  • 29ff5b8 [libtorch] [xnnpack] [qnnpack] [cpuinfo][fbgemm] (#46649)
  • 52d831c [libtorch] remove cuda from default features (#42917)
  • cbf6fe2 [opencv] add missing features to defaults and improve test matrix (#42371)
  • e1118ae [libtorch] Include <chrono> for high_resolution_clock (#43234)
  • 4413468 [sleef] Update to version 3.7 (#42706)
  • 8cbc1d7 [Libtorch] remove abs compiler path (#42729)
  • 88a0bf8 [onnx] Update onnx to 1.16.2 and fix downstream (#40067)
  • 6152303 [libtorch] Fix libtorch[core] compilation error (#40795)
  • fe1cde6 [fmt] update to 11.0.2 (#39738)
and 11 more commits:

# Source