Multi-Round Coreference Resolution (MRCR)

Randomized YaRN enhances LLM long-context reasoning

Researchers have developed Randomized YaRN, a novel training method designed to enhance the ability of large language models (LLMs) to generalize to significantly longer text sequences than they were initially trained o…