Module lora 

LoRA (Low-Rank Adaptation) for parameter-efficient fine-tuning.

Implements Hu et al. (2021): weight updates are decomposed as dW = B @ A where B in R^{d x r}, A in R^{r x k}, and r << min(d, k), drastically reducing trainable parameter count.

Re-exports

pub use adapter::LayerStats;

pub use adapter::LoraAdapter;

pub use adapter::LoraAdapterSummary;

pub use config::LoraConfig;

pub use error::LoraError;

pub use error::LoraResult;

pub use layer::LoraLayer;

Modules

adapter

Multi-layer LoRA adapter managing named LoRA layers.

config

LoRA configuration.

error

LoRA-specific error types.

layer

Core LoRA layer: low-rank A/B decomposition of a frozen base weight.