Struct LoadedLora 

pub struct LoadedLora {
    pub adapters: HashMap<String, Arc<LoraAdapter>>,
    pub rank: usize,
    pub alpha: f32,
}

A fully loaded LoRA adapter, mapping tensor base names to their adapters.

The key in adapters is the tensor base name without the .lora_a / .lora_b suffix (e.g. "blk.0.attn_q.weight").

After loading, attach adapters to QuantLinear layers via LoadedLora::get + QuantLinear::set_lora.

Fields

adapters: HashMap<String, Arc<LoraAdapter>>

Map from tensor base name → LoRA adapter.

rank: usize

The LoRA rank used throughout this adapter file.

alpha: f32

The alpha value (scale numerator).

Implementations

impl LoadedLora

pub fn load(path: &str) -> Result<LoadedLora, ArchError>

Load a LoRA adapter from a GGUF file on disk.

Errors

Returns ArchError::Gguf if the file cannot be parsed, or ArchError::Quant if dequantization of any adapter tensor fails.

pub fn from_gguf(model: &GgufModel) -> Result<LoadedLora, ArchError>

Load a LoRA adapter from an already-loaded GgufModel.

This is the primary construction path, separated from file I/O for testability.

pub fn get(&self, tensor_name: &str) -> Option<Arc<LoraAdapter>>

Look up the adapter for a named linear weight tensor.

Returns None if this LoRA file does not patch the named tensor.

pub fn num_adapters(&self) -> usize

Number of adapted layers in this adapter file.

Trait Implementations

impl Debug for LoadedLora

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

Formats the value using the given formatter.