Struct std::LoraConfig

#include <std/ml/lora.h> source

struct LoraConfig {
    unsigned long rank;
    float alpha;
    float dropout;
    unsigned long seed;
}

Fields

§unsigned long ranksource

r

§float alphasource

the update is scaled by alpha / r

§float dropoutsource

on the adapter's input while training (grad enabled)

§unsigned long seedsource

§static struct LoraConfig default(unsigned long rank)source · definition

alpha = rank (scale 1), no dropout.