DiaLong: Benchmarking Reminiscence in Long-Context Dialogues.

arXiv:2610.25463v2 Announce Type: replace-cross Abstract: How well can transformer-based Large Language Models (LLMs) retain and recall memories in long contexts of dialogue? LLMs are known to forget long-range factual information in their context, yet the fie