ALR$^2$: A Retrieve-then-Reason Framework for Long-context Question
Answering
ALR$^2$: A Retrieve-then-Reason Framework for Long-context Question
Answering
The context window of large language models (LLMs) has been extended significantly in recent years. However, while the context length that the LLM can process has grown, the capability of the model to accurately reason over that context degrades noticeably. This occurs because modern LLMs often become overwhelmed by the …