Expand description
§briny_ai
This crate provides a fast, minimal, and modular deep learning backend built in Rust. It features basic tensor operations, automatic differentiation, and support for CPU acceleration via Lazy SIMD. GPU support via WGPU is optional and designed for portability across Intel, AMD, and NVIDIA hardware.
§Features
- Tensors: N-dimensional arrays with shape tracking and gradient support
- Autograd: Functional-style forward and backward passes
- Operators: Efficient implementations of matrix multiplication,
ReLU, mean squared error, and stochastic gradient descent - GPU Acceleration: Optional
wgpu-powered compute shaders for matrix ops
Modules§
- approx
- Utilities to approximate equality of floating point values.
- backend
- Backend selection module.
- macros
- Provides the necessary means of abstraction which make advanced cases much simpler.
- nn
- Tedious manual tensor operations.
- prelude
- Common re-exports at a central location.
Macros§
- static_
model - Defines a deep learning model based off the descriptors.