LLM LORA configurations

The quantization range lies between -1 to 1 and uses 4-bit NormaFloat (NF4) and double quantization methods. QLoRA reduces the memory footprints using model parameters,