Windows binary build of the Torch machine learning framework
Clone or download
hiili Add more Visual Studio redistributable dlls
Copy <Visual Studio install directory>\VC\redist\x64\Microsoft.VC140.OpenMP\*.dll to bin\

Visual Studio version:
Microsoft Visual Studio Community 2015 Version 14.0.25431.01 Update 3
Latest commit b74692c Aug 1, 2017

Torch for Windows

This repository contains Windows binary builds of the Torch machine learning framework.

See the releases page or branches for 64-bit and 32-bit versions.

For more information, see and The process for producing these builds is documented here: The git logs contain the actual build commands and repository versions that were used.

Preinstalled packages

Preinstalled packages: cwrap, dok, fun, gnuplot, image, inspect, luaffi, luafilesystem, moses, nn, nnx, optim, paths, penlight, strict, sundown, sys, torch, xlua

The torch and nn packages are built from pinned versions (commit 7bbe179 on Sep 6, 2016, and commit 1b95f7e on Feb 21, 2017, respectively). Unfortunately, more recent versions do not compile on Windows when following the aforementioned build process.

Everything else is built from the most recent versions that were available when building (the build date is mentioned in the release titles).

Installing and running

  1. Choose a branch that contains the desired build configuration
  2. Download or clone the branch to C:\torch
  3. In a command prompt:
    1. C:\torch\setpaths.cmd
    2. luajit

If you wish to place the binaries somewhere else than C:\torch, then you need to modify the path configuration in several files (grep for 'C:[\\/]torch'). If you do not plan to use luarocks for installing additional packages, then simply modifying the paths in setpaths.cmd is enough. Torch itself works through symlinks and junctions, too, but luarocks doesn't.

If you want to plot from Torch using Gnuplot, then you need to have Gnuplot installed on your system and the bin directory of the installation in your PATH.

Local and remote debugging with ZeroBrane Studio should work for both 64-bit and 32-bit applications (see the torch7 wiki link above for details).


The lua51.dll file in the bin directory is simply a copy of libluajit.dll; this is needed for local debugging with ZeroBrane Studio.