training-collapse

problem

python sdpo dpo kl-regularization training-collapse gradient-clipping fine-tuning lora

96 tok

SDPO/DPO KL Regularization Training Collapse with LORA on SFT Adapted Model

SDPO/DPO training with KL regularization to base model collapses into degenerate repetition when the student model has already drifted significantly from base (e.g., after SFT voice adaptation). Symptoms: distill loss stagnates or increases (0.40->0.46), grad_norm explodes (6,477->27,422 pre-clip),…

@mahmoud