dfdx

[2] The only things that use Arc are tensors to store their data. Arc is used instead of Box to reduce allocations when tensors are cloned.

GPU acceleration with CUDA

Enable the cuda feature to start using the Cuda device! Requires the installation of nvidia's cuda toolkit. See feature flags docs for more info.

API Preview

Check examples/ for more details.

1. 👌 Simple Neural Networks API, completely shape checked at compile time.

type Mlp = (
    (Linear<10, 32>, ReLU),
    (Linear<32, 32>, ReLU),
    (Linear<32, 2>, Tanh),
);

fn main() {
    let dev: Cuda = Default::default(); // or `Cpu`
    let mlp = dev.build_module::<Mlp, f32>();
    let x: Tensor<Rank1<10>, f32, Cpu> = dev.zeros();
    let y: Tensor<Rank1<2>, f32, Cpu> = mlp.forward(x);
    mlp.save("checkpoint.npz")?;
}

1. 📈 Ergonomic Optimizer API

type Model = ...
let mut model = dev.build_module::<Model, f32>();
let mut grads = model.alloc_grads();
let mut sgd = Sgd::new(&model, SgdConfig {
    lr: 1e-2,
    momentum: Some(Momentum::Nesterov(0.9))
});

let loss = ...
grads = loss.backward();

sgd.update(&mut model, &grads);

1. 💡 Const tensors can be converted to and from normal rust arrays

let t0: Tensor<Rank0, f32, _> = dev.tensor(0.0);
assert_eq!(t0.array(), &0.0);

let t1 /*: Tensor<Rank1<3>, f32, _>*/ = dev.tensor([1.0, 2.0, 3.0]);
assert_eq!(t1.array(), [1.0, 2.0, 3.0]);

let t2: Tensor<Rank2<2, 3>, f32, _> = dev.sample_normal();
assert_ne!(t2.array(), [[0.0; 3]; 2]);

Fun/notable implementation details

Module

pub trait Module<Input> {
    type Output;
    fn forward(&self, input: Input) -> Self::Output;
}

From this flexible trait we get:

1. Single & batched inputs (just have multiple impls!)
2. Multiple inputs/outputs (multi-headed modules, or rnns)
3. Behavior different when tape is present or not (not the .train()/.eval() behavior present in other libraries!).

Tuples represent feedforward (a.k.a sequential) modules

Since we can implement traits for tuples, which is not possible in other languages AFAIK, they provide a very nice frontend for sequentially executing modules.

// no idea why you would do this, but you could!
type Model = (ReLU, Sigmoid, Tanh);
let model = dev.build_module::<Model, f32>();

type Model = (Linear<10, 5>, Tanh)
let model = dev.build_module::<Model, f32>();

How implementing Module for a 2-tuple looks:

impl<Input, A, B> Module<Input> for (A, B)
where
    Input: Tensor,
    A: Module<Input>,        // A is a module that takes Input
    B: Module<A::Output>,    // B is a module that takes A's Output
{
    type Output = B::Output; // the output of this is B's Output
    fn forward(&self, x: Input) -> Self::Output {
        let x = self.0.forward(x);
        let x = self.1.forward(x);
        x
    }
}

Modules implemented for Tuples up to 6 elements, but you can arbitrarily nest them!

No Rc<RefCells<T>> used - Gradient tape is not kept behind a cell!

Other implementations may store a reference to the gradient tape directly on tensors, which requires mutating tensors or using Rc/Refcells all over the place.

We've figured out an elegant way to avoid this, reducing references and dynamic borrow checks to 0!

Since all operations result in exactly 1 child, we can always move the gradient tape to the child of the last operation. Additionally, no model parameters (all tensors) will ever own the gradient tape because they will never be the result of any operation. This means we know exactly which tensor owns the gradient tape, and the tensors that have it will always be intermediate results that don't need to be maintained across gradient computation.

All of this together gives users unprecedented control/precision over what tensors are recorded on the gradient tape!

One advanced use case requires that tensors be re-used multiple times in a computation graph. This can be handled by cloning the tensor, and manually moving the gradient tape around.

Type checked backward

tl;dr: If you forget to include a call to trace() or traced(), the program won't compile!

-let pred = module.forward(x);
+let pred = module.forward(x.traced(grads));
let loss = (y - pred).square().mean();
let gradients = loss.backward();

Since we know exactly what tensors own the gradient tape, we can require the tensor passed into .backward() to own the gradient tape! And further, we can require it be moved into .backward(), so it can destruct the tape and construct the gradients!

All of this can be checked at compile time 🎉

📄 Validated against pytorch

All functions & operations are tested against behavior shown by similar code in pytorch.

License

Dual-licensed to be compatible with the Rust project.

Licensed under the Apache License, Version 2.0 http://www.apache.org/licenses/LICENSE-2.0 or the MIT license http://opensource.org/licenses/MIT, at your option. This file may not be copied, modified, or distributed except according to those terms.