Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings - Sakana AI

Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings Sakana AI