viva_tensor/layout

Tensor layout metadata.

A tensor is not just its values. To interpret tensor storage correctly, every backend needs the same small contract: shape, strides, offset, dtype, device, and storage representation.

Types

TensorDevice

</>

Where the tensor payload lives.

pub type TensorDevice {
  BeamCpu
  NativeCpu
  CudaDevice(Int)
}

Constructors

```
BeamCpu
```
```
NativeCpu
```
```
CudaDevice(Int)
```

TensorDtype

</>

Element type for this tensor value.

pub type TensorDtype {
  Float64
  Float32
  Float16
  BFloat16
  Float8E4M3
  Int8
  Int4
  SparseFloat16
}

Constructors

```
Float64
```
```
Float32
```
```
Float16
```
```
BFloat16
```
```
Float8E4M3
```
```
Int8
```
```
Int4
```
```
SparseFloat16
```

TensorLayout

</>

Canonical metadata for interpreting tensor storage.

pub type TensorLayout {
  TensorLayout(
    storage: TensorStorage,
    device: TensorDevice,
    dtype: TensorDtype,
    shape: List(Int),
    strides: List(Int),
    offset: Int,
    size: Int,
    rank: Int,
    contiguous: Bool,
  )
}

Constructors

TensorLayout(
  storage: TensorStorage,
  device: TensorDevice,
  dtype: TensorDtype,
  shape: List(Int),
  strides: List(Int),
  offset: Int,
  size: Int,
  rank: Int,
  contiguous: Bool,
)

TensorMemoryLayout

</>

Logical memory layout used by runtime planning.

pub type TensorMemoryLayout {
  RowMajor
  ColumnMajor
  StridedLayout
  PackedFp8Layout
  PackedSparse24Layout
}

Constructors

```
RowMajor
```
```
ColumnMajor
```
```
StridedLayout
```
```
PackedFp8Layout
```
```
PackedSparse24Layout
```

TensorStorage

</>

How the tensor’s payload is represented.

pub type TensorStorage {
  DenseStorage
  StridedStorage
  NativeStorage
}

Constructors

```
DenseStorage
```
```
StridedStorage
```
```
NativeStorage
```