Neurox is a fast, minimalist, and extendable numerical computation & machine learning library written in Rust.
It provides tensor operations, activation functions, layer abstractions, and model building blocks to create and run ML models.
Currently optimized for CPU execution, with a GPU backend planned in future releases.
Perfect for:
- Learning how ML frameworks work under the hood
- Building lightweight ML models in Rust
- Using as a base for larger GPU-accelerated projects
- Multi-dimensional Tensor struct for efficient numerical storage
- Matrix operations: multiplication, addition, dot products
- Activation functions: ReLU, Sigmoid, Tanh, Softmax, LeakyReLU, XOR-like logical ops
- Layer system: Dense layers with bias & activation support
- Model API: Create, add layers, run forward passes
- Logical / Boolean operations on tensors (e.g., XOR, AND, OR)
- Random initialization utilities
- Device abstraction for future GPU acceleration
- Modular architecture extend with custom layers or activations easily
- Example scripts for quick usage
Add Neurox to your Rust project:
[dependencies]
neurox = "0.2.0"