Enum GgmlType 

pub enum GgmlType {
    F32,
    F16,
    Bf16,
    Q4_0,
    Q4_1,
    Q5_0,
    Q5_1,
    Q8_0,
    Q8_1,
    Q2K,
    Q3K,
    Q4K,
    Q5K,
    Q6K,
    Q8K,
    Tq1_0,
    Tq2_0,
    Q2_0 {
        group: usize,
    },
}

A GGML block-quantization type.

The PrismML-private Q2_0 (ggml id 42) is polymorphic in its group size (128 in the shipped Ternary-Bonsai file, 64 in the fork master), so the group is carried in the variant rather than hardcoded. Standard Tq2_0 (id 35, group-256, scale-at-end) is a separate entry and must not be conflated with it.

Variants

F32

F16

Bf16

Q4_0

Q4_1

Q5_0

Q5_1

Q8_0

Q8_1

Q2K

Q3K

Q4K

Q5K

Q6K

Q8K

Tq1_0

Tq2_0

Q2_0

PrismML id 42: 2-bit, symmetric, scale-at-front, contiguous codes. group ∈ {64, 128}; block = 2 + group/4 bytes.

Fields

group: usize

Implementations

impl GgmlType

pub fn from_id(id: u32, q2_0_group: usize) -> Result<Self>

Map a raw ggml type id to a GgmlType. q2_0_group supplies the group size for the id-42 PrismML type (ignored for every other id); pass the value read from the fork’s GGUF metadata (128 default, 64 fork-master).

pub fn block_size(self) -> usize

Number of dequantized elements produced by one block (QK / QK_K).

pub fn block_bytes(self) -> Result<usize>

On-disk byte size of one block.

Trait Implementations

impl Clone for GgmlType

fn clone(&self) -> GgmlType

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GgmlType

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

Formats the value using the given formatter.

impl PartialEq for GgmlType

fn eq(&self, other: &GgmlType) -> 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 Copy for GgmlType

impl Eq for GgmlType

impl StructuralPartialEq for GgmlType