In Search of Needles in a 11M Haystack: Recurrent Memory Finds What LLMs Miss
This addresses the challenge of extracting distributed facts in extensive texts for AI applications, representing a significant rather than incremental advance in long-sequence processing.
The paper tackles the problem of processing long documents with generative transformer models by introducing the BABILong benchmark and showing that fine-tuning GPT-2 with recurrent memory augmentations enables handling tasks with up to 11 million elements, a substantial leap over existing methods limited to 10,000 elements.
This paper addresses the challenge of processing long documents using generative transformer models. To evaluate different approaches, we introduce BABILong, a new benchmark designed to assess model capabilities in extracting and processing distributed facts within extensive texts. Our evaluation, which includes benchmarks for GPT-4 and RAG, reveals that common methods are effective only for sequences up to $10^4$ elements. In contrast, fine-tuning GPT-2 with recurrent memory augmentations enables it to handle tasks involving up to $11\times 10^6$ elements. This achievement marks a substantial leap, as it is by far the longest input processed by any neural network model to date, demonstrating a significant improvement in the processing capabilities for long sequences.