EpochLex entry
QLoRA
cue-LOHR-uh
A parameter-efficient fine-tuning technique that keeps a pretrained model quantized, commonly at 4-bit precision, while training low-rank adapter weights instead of updating the full model.
EpochLex entry
cue-LOHR-uh
A parameter-efficient fine-tuning technique that keeps a pretrained model quantized, commonly at 4-bit precision, while training low-rank adapter weights instead of updating the full model.