laser

The Truth Is In There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction