Understanding LoRA Rank Trade-offs in Diffusion Model Fine-Tuning
Selecting LoRA rank for diffusion fine-tuning requires balancing quality and compute cost. We present a controlled study on CIFAR-10 using a DDPM U-Net with ranks {2,4,8,16,32}, fixed optimization settings, and a reproducible local-folder pytorch-fid protocol. We report FID, trainable parameters, runtime, and GPU memor...
Iman Khazrak, N. Nejad, Mostafa M. Rezaee et al.
· 0 citations