Scale-QLoRA: Code-Invariant Adapter Merging for Native 4-bit Microscaling LLMs
Scale-QLoRA and merge-aware QAT-LoRA are both accuracy-lossless, so they claim no accuracy ordering between them; they differ structurally, in that QAT-LoRA re-derives the code plane through a quantizer while Scale-QLoRA preserves it exactly.
Tung-Ling Li, Jia-Le Huang, Lee-Chi Wang et al.
· 1 citation