Struct LlamaQuantizeParams 

pub struct LlamaQuantizeParams { /* private fields */ }

Parameters controlling a quantize run.

Starts from llama_model_quantize_default_params(); the builder methods override the handful of fields common quantization runs need. ik’s per-tensor override fields (output_tensor_type, token_embedding_type, attn_q_type, attn_k_type, attn_v_type, attn_qkv_type, attn_output_type, ffn_gate_type, ffn_down_type, ffn_up_type, ffn_gate_inp_type, extra_output_type, per_layer_token_embedding_type) and the extra flags (only_copy, pure_, keep_split, ignore_imatrix_rules, only_repack, dry_run, partial_requant) plus the opaque pointers (imatrix, kv_overrides, custom_quants, repack_pattern, user_data) keep their default values unless set through as_raw_mut.

Implementations

impl LlamaQuantizeParams

pub fn with_ftype(self, ftype: llama_ftype) -> Self

Target quantization format (e.g. ftype::LLAMA_FTYPE_MOSTLY_Q4_K_M or an ik SOTA quant like ftype::LLAMA_FTYPE_MOSTLY_IQ4_K).

pub fn with_n_threads(self, n_threads: i32) -> Self

Number of threads to use (0 lets the C side pick a default).

pub fn with_allow_requantize(self, allow_requantize: bool) -> Self

Allow requantizing tensors that are already quantized (default false).

pub fn with_quantize_output_tensor(self, quantize_output_tensor: bool) -> Self

Also quantize the output.weight tensor (default false keeps it larger for quality).

pub fn as_raw(&self) -> &llama_model_quantize_params

Immutable access to the raw params (advanced/escape hatch).

pub fn as_raw_mut(&mut self) -> &mut llama_model_quantize_params

Mutable access to the raw params, for setting ik’s extended per-tensor override fields, flags, or the opaque imatrix/kv_overrides pointers that are not surfaced by dedicated builders.

Trait Implementations

impl Clone for LlamaQuantizeParams

fn clone(&self) -> LlamaQuantizeParams

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LlamaQuantizeParams

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

Formats the value using the given formatter.

impl Default for LlamaQuantizeParams

fn default() -> Self

Returns the “default value” for a type.