On Reinforcement Learning and Distribution Matching for Fine-Tuning Language Models with no Catastrophic Forgetting. (arXiv:2206.00761v1 [cs.LG]) http://arxiv.org/abs/2206.00761