Enum TensorEncoding 

pub enum TensorEncoding {
    Dense {
        tensor: StoredTensor,
    },
    Affine {
        bits: u8,
        group_size: u64,
        group_axis: usize,
        packing: BitPacking,
        codes: StoredTensor,
        scales: StoredTensor,
        offset: AffineOffset,
    },
    Ggml {
        encoding: GgmlEncoding,
        data: DataSpan,
    },
    Opaque {
        name: String,
        metadata: Value,
        data: Vec<DataSpan>,
    },
}

Storage layout and quantization metadata, with explicit references to all data.

Variants

Dense

Addressable scalar values with explicit shape and strides.

Fields

tensor: StoredTensor

Physical view of the scalar values.

Affine

Grouped integer codes reconstructed using scales and an affine offset.

Fields

bits: u8

Bits per quantized value.

group_size: u64

Logical values sharing one scale and offset.

group_axis: usize

Axis along which quantization groups are formed.

packing: BitPacking

Order of codes within each storage word.

codes: StoredTensor

Packed integer codes.

scales: StoredTensor

One scale per quantization group.

offset: AffineOffset

Additive biases or integer zero points, one per group.

Ggml

GGML blocks whose layout is determined by the encoding code.

Fields

encoding: GgmlEncoding

data: DataSpan

Opaque

Uninterpreted storage retained for a future adapter.

Fields

name: String

metadata: Value

data: Vec<DataSpan>

Implementations

impl TensorEncoding

pub fn data(&self) -> Vec<&DataSpan>

Explicit references remain enumerable even for an unknown encoding.

impl TensorEncoding

pub fn validate(&self) -> Result<(), Error>

Check component layouts and supported affine grouping constraints. For GGML and opaque data, only span arithmetic is checked.

Trait Implementations

impl Clone for TensorEncoding

fn clone(&self) -> TensorEncoding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TensorEncoding

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for TensorEncoding

fn deserialize<__D>( __deserializer: __D, ) -> Result<TensorEncoding, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for TensorEncoding

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.