Crate ruda_tensor 

This library provides the core types that define how Ruda tensor data is represented, stored, and interpreted.

Re-exports

pub use backend::*;

pub use shape::*;

pub use slice::*;

pub use indexing::*;

Modules

apiapi

Public tensor API built on the shared backend contracts and primitives.

backend

Backend trait and required types.

cast

Tensor element casting.

collective

Explicit rank communicators for floating tensor collectives.

device

distribution

Random value distributions used to initialize and populate tensor data.

element

Traits and helpers for working with element types and conversions.

expert_projection

Original native trainable floating expert projections and source SwiGLU derivatives. Original native floating expert projections and storage-rounded SwiGLU input VJPs.

frozen_awq

Native frozen AWQ projections and their packed-weight input derivatives. Optional backend extension for frozen, directly packed AWQ projections.

frozen_nf4

Original RUDA byte-packed NF4 frozen projections and input derivatives. Backend extension for the original RUDA high-nibble-first NF4 format.

grouped_nf4

Original byte-packed NF4 selected expert projections and SwiGLU derivatives. Frozen original RUDA NF4 expert cubes, with discrete native U32 row assignments.

indexing

Indexing utilities.

moe

Native selected router weights and complete local MoE training operations. Native local MoE training contracts over existing device routing and expert kernels.

moe_exchange

Original native expert-parallel dispatch/receive/combine contracts. Native routing/dispatch, received expert rows and ordered combine for expert-parallel graphs.

ops

packed_experts

Original independently selected AWQ/NF4 expert payloads and native derivatives. Actual original AWQ/NF4 selected expert payloads and their native first-order chains.

primitive

quantization

Quantization data representation.

shape

Shape definition.

slice

Slice utilities.

tensor

Backend tensor primitives and operations.

Macros

dequant_op_flow

Automatically applies dequantization -> float operation [-> quantization].

dequant_op_quant

Automatically applies dequantization -> float operation -> quantization.

doc_tensor

Convenience macro to link to the ruda-tensor API docs.

make_element

Macro to implement the element trait for a type.

Structs

Bytes

A buffer similar to Box<[u8]> that supports custom memory alignment and allows trailing uninitialized bytes.

DeviceHandle

TODO: Docs

DistributionSampler

Distribution sampler for random value of a tensor.

StreamId

Unique identifier that can represent a stream based on the current thread id.

TensorData

Data structure for tensors.

Tolerance

The tolerance used to compare to floating point numbers.

bf16

A 16-bit floating point type implementing the bfloat16 format.

f16

A 16-bit floating point type implementing the IEEE 754-2008 standard binary16 a.k.a “half” format.

Enums

AllocationProperty

The kind of allocation behind the Bytes type.

BoolStore

Data type used to store boolean values.

DType

DataError

The things that can go wrong when manipulating tensor data.

Distribution

Distribution for random value of a tensor.

DistributionSamplerKind

Distribution sampler kind for random value of a tensor.

FloatDType

IntDType

Scalar

A scalar element.

Traits

Element

Core element trait for tensor values.

ElementComparison

Element ordering trait.

ElementConversion

Element conversion trait for tensor.

ElementEq

Element trait for equality of a tensor.

ElementLimits

Element limits trait.

ElementOrdered

Ordered element trait for tensor values.

ElementRandom

Element trait for random value of a tensor.

Functions

read_sync

Read a future synchronously.

try_read_sync

Read a future synchronously.

Type Aliases

BoolDType

Boolean dtype.