Understanding LLM Performance Degradation in Multi-Instance Processing: The Roles of Instance Count and Context Length

arXiv:2603.22608v2 Announce Type: replace Abstract: Users often rely on Large Language Models (LLMs) for processing multiple documents or performing analysis over a number of instances. For example, analysing the overall sentiment of a number of movie reviews requires an LLM to process the...