Llama — Field Evidence

For What Reason? Interpreting Models' Encoding of Causation and Antithesis — Discourse relations provide document structure, critical to language understanding and enabling language model performance and ethicality. In this work, we investigate how instruction-tuned Transformer m