luma-optim

Optimizers (SGD, SGDM, Momentum, RMSProp, Adam, AdamW) for updating gradients on any luma-tensor device.

Usage

use luma_nn::{Linear, Module};
use luma_optim::{Optimizer, SGD};
use luma_tensor::Cpu;

let model = Linear::new(3, 4, true, Cpu).unwrap();
let mut optimizer = SGD::new(model.params(), 0.01);

// after a forward pass and `let grads = loss.backward()?`:
// optimizer.step(&grads)?;

Adam

use luma_optim::{Adam, AdamConfig};

let optimizer = Adam::new(params, AdamConfig::default()).unwrap();

License

MIT — see LICENSE.