DroPE: Extending the Context of LLMs by Dropping Their Positional Embeddings

(pub.sakana.ai) by hardmaru | Jan 12, 2026 | 0 comments on HN