Adapt large models with LoRA, adapters, prompt tuning, and quantized low-rank training. Choose target layers and ranks, merge adapters, and compare quality, memory use, and deployment constraints.
