flashlight_tensor

Tensor library written in pure rust, designed mostly for matrix operations

Earlier I decided to abandon the project, but I guess, I still want to work on that project

project not related to similarly named flashlight. The name was coincidental and chosen independently.

Features

• n-dimensional tensors
• Element-wise operations
• Scalar multiplication and addition
• Tensor multiplication and addition
• Matrix transformation
• Dot product
• ReLU and sigmoid
• CPU only, with partial GPU support

Instalation

[dependencies]
flashlight_tensor = "0.2.7"

// Experimental
flashlight_tensor = { git = "https://github.com/Bejmach/flashlight_tensor"}

Documentation

Docs

all tensor operations in tensor category

Quick Start

For gpu usage go to examples on github

use flashlight_tensor::prelude::*;

fn main(){
    //2 rows, 2 collumns, fill with 1.0
    let a: Tensor<f32> = Tensor::fill(1.0, &[2, 2]);
}

Tests

Run tests with:
cargo test

G# means github version # of patch. You can treat that as alpha patch of next version. Versions G are avilable on github early, and those features will be released on cargo with next bigger patch. Not everything in G version is tested and working. You are using it at your own responsibility.

Patch notes

• V0.2.4:
  • matrix_vec/col, now return a matrix, not vector
  • matrix_col/row_sum/prod, return a sum/product of all collumns/rows in matrix
• V0.2.5
  • G1
    • mutable operations for iterative functions
  • G2
    • better file structure
• V0.2.6
  • activation functions for neural network
• V0.2.6
  • G1
    • wgpu preparation, currently not working
  • G2
    • gpu_buffers that allows for running tensor operations on gpu, for now only addition, and cpu preparation unoptimized
  • G3
    • first gpu operations and tests, and matmul gpu vs cpu comparison in examples
  • G4
    • most operations and tests on gpu, No docs for now
• V0.3.0 - most operations + basic docs

Plans for 0.4.0?

• Good question, but currently no