Crate axonml_tensor 

N-dimensional tensor library for AxonML.

Exports Tensor<T> (generic over Scalar types) with NumPy-style broadcasting, strided zero-copy views, CPU + CUDA GPU matmul (with GEMV fast path for m=1 decode), quantized matmul dispatch (Q4_K/Q6_K in-shader dequant via cuda_ops), lazy tensors with algebraic optimization, sparse COO tensors, factory functions (zeros/ones/randn/arange/linspace/eye/full), and shape/stride utilities. Re-exports Device, DType, Error, Result from axonml-core.

Modules: tensor, shape, creation, ops, view, cuda_ops, lazy, sparse.

File

crates/axonml-tensor/src/lib.rs

Author

Andrew Jewell Sr. — AutomataNexus LLC ORCID: 0009-0005-2158-7060

Updated

April 14, 2026 11:15 PM EST

Disclaimer

Use at own risk. This software is provided “as is”, without warranty of any kind, express or implied. The author and AutomataNexus shall not be held liable for any damages arising from the use of this software.

Re-exports

pub use lazy::LazyOp;

pub use lazy::LazyTensor;

pub use shape::Shape;

pub use shape::Strides;

pub use tensor::Tensor;

pub use creation::*;

Modules

creation

Tensor factory functions.

lazy

Lazy tensors — deferred computation with algebraic graph optimization.

ops

Higher-level tensor operations — activations, comparisons, clamping.

prelude

Convenient imports for common usage.

shape

Shape and stride utilities for tensor dimension management.

sparse

Sparse tensor support — COO, CSR, CSC formats.

tensor

Core Tensor<T> struct — 2170 lines, 76 public methods.

view

Views, slicing, and advanced indexing operations on Tensor<T>.

Enums

DType

Runtime representation of tensor data types.

Device

Represents a compute device where tensors can be allocated and operations executed.

Error

The main error type for Axonml operations.

Type Aliases

Result

A specialized Result type for Axonml operations.