dkappe / ThArrays.jl

A Julia interface for PyTorch's C++ backend, focusing on Tensor, AD, and JIT

ThArrays

A Julia interface for PyTorch's C++ backend.

Features

• ThArrays.Tensor: PyTorch Tensor as an Array-like data type in Julia
• ThArrays.ThAD: AD using PyTorch C++ backend
• ThArrays.TrackerAD: AD using Tracker.jl and PyTorch C++ backend mixed, on your choice
• ThArrays.ThJIT: using TorchScript in Julia

Getting Started

1. Install the package: ] add ThArrays

2. Read the docs here, or

3. Experiment in the Julia REPL directly:

    julia> using ThArrays
   
    julia> t = Tensor( -rand(3, 3) )
    PyTorch.Tensor{Float64, 2}:
    -0.1428 -0.7099 -0.1446
    -0.3447 -0.0686 -0.8287
    -0.2692 -0.0501 -0.2092
    [ CPUDoubleType{3,3} ]
   
    julia> sin(t)^2 + cos(t)^2
    PyTorch.Tensor{Float64, 2}:
     1.0000  1.0000  1.0000
     1.0000  1.0000  1.0000
     1.0000  1.0000  1.0000
    [ CPUDoubleType{3,3} ]
   
    julia> ThAD.gradient(x->sum(sin(x)+x^2), rand(3,3))
    (PyTorch.Tensor{Float64, 2}:
     2.3776  1.5465  2.0206
     1.2542  1.2081  2.1156
     2.1034  1.1568  2.2599
    [ CPUDoubleType{3,3} ]
    ,)
   
    julia>
   

   You can find more examples under the test directory.