1 paper · 1 filter
Sikui Zhang, Guangze Gao, Ziyun Gan +5
Large language models (LLMs) experience significant performance degradation when the input exceeds the pretraining context window, primarily due to the out-of-distribution (OOD) be…