Memory Tokens: Large Language Models Can Generate Reversible Sentence Embeddings
This work addresses the need for memory-based retrieval and compression in AI, though it appears incremental as it builds on existing LLM capabilities without major methodological shifts.
The researchers tackled the problem of generating reversible sentence embeddings that allow large language models to reconstruct original text exactly, achieving perfect reconstruction on sequences up to 240 tokens with models like Llama 3.1 8B.
In this work, we observe an interesting phenomenon: it is possible to generate reversible sentence embeddings that allow an LLM to reconstruct the original text exactly, without modifying the model's weights. This is achieved by introducing a special memory token, whose embedding is optimized through training on a fixed sequence. When prompted with this embedding, the model reconstructs the fixed sequence exactly. We evaluate this phenomenon across English and Spanish datasets, sequences of up to approximately 240 tokens, and model scales ranging from 100M to 8B parameters. Notably, Llama 3.1 8B successfully reconstructs all tested sequences. Our findings highlight an interesting capability of LLMs and suggest potential applications in memory-based retrieval, compression, and controlled text generation.