Interpretable Catastrophic Forgetting of Large Language Model Fine-tuning via Instruction Vector

Open in new window