Struct FitParamsResult 

pub struct FitParamsResult {
    pub model_params: LlamaModelParams,
    pub context_params: LlamaContextParams,
    pub tensor_split: Vec<f32>,
    pub margins: Vec<usize>,
    /* private fields */
}

Fitted model/context parameters plus auxiliary buffers.

LlamaModelParams inside this struct may point at Self::tensor_split and its internal tensor buffer-type override storage; keep the whole FitParamsResult alive while loading a model with these parameters.

Fields

model_params: LlamaModelParams

Model parameters after fitting (n_gpu_layers, tensor split, …).

context_params: LlamaContextParams

Context parameters after fitting (n_ctx, …).

tensor_split: Vec<f32>

Layer split ratios per device (writable buffer passed to llama.cpp).

margins: Vec<usize>

Per-device memory margins used during fitting (bytes).

Implementations

impl FitParamsResult

pub fn active_tensor_split(&self) -> &[f32]

Tensor split values for active devices, trimming trailing zeros.

Trait Implementations

impl Debug for FitParamsResult

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

Formats the value using the given formatter.