LoRA, QLoRA, VeRA and more: what each of 6 fine-tuning variants actually optimizes

techNmak · x · 2026-09-20

Everyone fine-tunes LLMs, but few understand what's actually being updated. LoRA, QLoRA, LoRA-FA, VeRA, Delta-LoRA and LoRA+ are not small variations of one method—each attacks a different bottleneck.

Original post →

More from Research

Research channel →