ferrotorch-core

Core tensor and autograd engine for ferrotorch -- PyTorch in Rust.

What it provides

• Tensor<T> -- N-dimensional tensor with automatic differentiation
• Autograd -- reverse-mode automatic differentiation via backward(), no_grad(), and autocast
• Creation functions -- tensor, zeros, ones, rand, randn, arange, linspace, eye, full, scalar, zeros_like, ones_like, rand_like, randn_like, full_like
• Arithmetic & math ops -- element-wise operations with broadcast support, differentiable exp, log, sin, cos, clamp
• Tensor manipulation -- permute, view, contiguous, chunk, split
• Reductions -- sum_dim, mean_dim with axis and keepdim support
• Einsum -- differentiable Einstein summation (einsum, einsum_differentiable)
• FFT -- fft, ifft, rfft, irfft, fft2, ifft2 with differentiable variants
• Quantization -- quantize, dequantize, quantized_matmul, QuantizedTensor
• Sparse tensors -- SparseTensor for COO-format sparse data
• Storage -- TensorStorage, StorageBuffer for raw memory management
• DType system -- DType, Element, Float traits covering f32, f64, f16, bf16
• Device abstraction -- Device enum for CPU/GPU placement

Quick start

use ferrotorch_core::{tensor, backward, no_grad, Tensor, Float};

fn main() {
    let x = tensor(&[1.0_f32, 2.0, 3.0]).requires_grad(true);
    let y = (&x * &x).sum(None);
    backward(&y).unwrap();
    println!("grad: {:?}", x.grad().unwrap());
}

Part of ferrotorch

This crate is one component of the ferrotorch workspace. See the workspace README for full documentation.

License

MIT OR Apache-2.0