Struct OneBitTensor 

pub struct OneBitTensor<'a> {
    pub name: String,
    pub shape: Vec<u64>,
    /* private fields */
}

A 1-bit tensor backed by Q1_0_g128 blocks.

This wraps raw GGUF tensor data and provides typed access to blocks without copying or dequantizing the entire tensor.

Fields

name: String

Tensor name.

shape: Vec<u64>

Shape dimensions.

Implementations

impl<'a> OneBitTensor<'a>

pub fn from_raw( name: String, shape: Vec<u64>, data: &'a [u8], ) -> BonsaiResult<Self>

Create a 1-bit tensor from raw GGUF tensor data bytes.

pub fn num_blocks(&self) -> usize

Number of blocks in this tensor.

pub fn element_count(&self) -> usize

Total number of elements (weights) in this tensor.

pub fn block(&self, index: usize) -> &BlockQ1_0G128

Get a reference to the block at the given index.

pub fn blocks(&self) -> &[BlockQ1_0G128]

Get all blocks as a slice.

pub fn dequantize_all(&self) -> Vec<f32>

Dequantize all blocks to FP32 values.

For the full tensor, this allocates and fills an output vector. For per-operation dequantization, use the kernel crate instead.

Trait Implementations

impl<'a> Debug for OneBitTensor<'a>

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

Formats the value using the given formatter.