Stabilizing language models under continual learning via condition-anchored distillation
Condition-anchored functional preservation as a common design principle across the tested language-generation objectives is supported, and the implementation directly controls local denoising drift on teacher-generated completions.