The Fragile Truth of Saliency: Improving LLM Input Attribution via Attention Bias Optimization

Open in new window