Stabilizing language models under continual learning via condition-anchored distillation

Excerpt

arXiv:2610.06940v1 Announce Type: new Abstract: Continual adaptation of language models can change their output distribution on prompts learned earlier, while retaining every old prompt-answer pair may be undesirable or impossible. We study condition-anchored generative distillation (CAGD): retain a small set of old prompts, use a frozen previous model to reconstruct completions and generation states, and match its predictive distributions while learning the next task. The formulation separates