Struct TensorTable 

pub struct TensorTable {
    pub blocks: u64,
    pub tensors: Vec<TensorEntry>,
}

A checkpoint’s tensor table and the blocks its step runs (the rest are drafting blocks).

Fields

blocks: u64

tensors: Vec<TensorEntry>

Implementations

impl TensorTable

pub fn from_rows( blocks: u64, rows: impl IntoIterator<Item = (String, Vec<u64>, u32)>, ) -> Result<TensorTable, ImageError>

The table of rows (name, dims, GGUF type id), the first blocks blocks being the step’s.

pub fn of_gguf(g: &Gguf) -> Result<TensorTable, ImageError>

The table of a checkpoint on disk.

pub fn of_header(h: &GgufHeader) -> Result<TensorTable, ImageError>

The table of a checkpoint’s header ledger.

pub fn read_order(&self) -> Vec<usize>

The order a pass reads the tensors: the embedding table, each of the step’s blocks as BLOCK_READS lists it, the output norm and the head (a head tied to the embedding table reads the table already placed), then every tensor the step does not read, in the table’s order.

Trait Implementations

impl Clone for TensorTable

fn clone(&self) -> TensorTable

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorTable

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

Formats the value using the given formatter.

impl PartialEq for TensorTable

fn eq(&self, other: &TensorTable) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for TensorTable

impl StructuralPartialEq for TensorTable