Poro is a library I am writing to help me better understand modern ML frameworks like Pytorch and Tensorflow. It is mostly based off of Karpathys Micrograd series of (lectures)[https://www.youtube.com/watch?v=VMj-3S1tku0]. It is in rust because I enjoy its ease of setup.
- Basic neural network operations
- Frictionless Autograd
- Cuda support (limited by growing!)
- Support for custom layers and operations
- Lightweight and trying to focus on ease of understanding for others
I work on this lib while I also have a job, and it is done for the enjoyment of learning, so it is likely a good idea to not use this for professional use as I am unlikely to get to any issues you might have in a timely manner :)
if you run cargo test, some tests might fail, this is not actually them failing do to the nature of the way I have the equation working as a singelton, it it possible for test to grab the lock at the wrong time and fail. If you know how to fix this, that would make a great first PR :). You can always run the tests one at a time and see that they work that way
cargo test --releaseTo use Poro, add the following to your Cargo.toml:
[dependencies]
poro = "0.1.2"Here is a simple example to get you started with Poro:
use Poro::tensor::Tensor;
use ndarray::prelude::*;
fn main() {
let a = Tensor::ones(Shape::new(vec![2, 2]));
let b = Tensor::zeroes(Shape::new(vec![2, 2]));
let c = a + b;
let result = c.item();
assert!(result == arr2(&[[1.0, 1.0], [1.0, 1.0]]).into_dyn());
}- Optimizer module
- Data Loader Module
- Working with Metal
- Transfomers Layer
- Conv Layer
- Model/Module Configure