Needle in the Haystack for Memory Based Large Language Models
This addresses memory limitations in LLMs for long-context recall tasks, offering a more efficient solution compared to larger or modified models, though it is incremental as it builds on existing memory-based architectures.
The paper tackled poor fact retrieval in large language models by coupling an external associative memory to Larimar, enabling it to handle contexts much longer than training data without task-specific training, with a relatively smaller model maintaining strong performance.
Current large language models (LLMs) often perform poorly on simple fact retrieval tasks. Here we investigate if coupling a dynamically adaptable external memory to a LLM can alleviate this problem. For this purpose, we test Larimar, a recently proposed language model architecture which uses an external associative memory, on long-context recall tasks including passkey and needle-in-the-haystack tests. We demonstrate that the external memory of Larimar, which allows fast write and read of an episode of text samples, can be used at test time to handle contexts much longer than those seen during training. We further show that the latent readouts from the memory (to which long contexts are written) control the decoder towards generating correct outputs, with the memory stored off of the GPU. Compared to existing transformer-based LLM architectures for long-context recall tasks that use larger parameter counts or modified attention mechanisms, a relatively smaller size Larimar is able to maintain strong performance without any task-specific training or training on longer contexts.