Rep2Text: Decoding Full Text from a Single LLM Token Representation

TL;DR

This paper introduces Rep2Text, a framework that decodes input text from a single last-token representation in LLMs, revealing information bottlenecks and generalization capabilities.

Contribution

Rep2Text is a novel method that reconstructs input text from last-token representations, highlighting the extent of information retention and semantic preservation in LLMs.

Findings

Approximately half of tokens in 16-token sequences can be recovered.

Token recovery declines with increasing sequence length.

The framework generalizes well to out-of-distribution clinical data.

Abstract

Large language models (LLMs) have achieved remarkable progress across diverse tasks, yet their internal mechanisms remain largely opaque. In this work, we investigate a fundamental question: to what extent can the original input text be recovered from a single last-token representation in an LLM? To this end, we propose Rep2Text, a novel framework for decoding text from last-token representations. Rep2Text employs a trainable adapter that maps a target model's last-token representation into the token embedding space of a decoding language model, which then autoregressively reconstructs the input text. Experiments across various model combinations (Llama-3.1-8B, Gemma-7B, Mistral-7B-v0.1, Llama-3.2-3B, etc.) show that, on average, roughly half of the tokens in 16-token sequences can be recovered from this compressed representation while preserving strong semantic coherence. Further analysis reveals a clear information bottleneck effect: as sequence length increases, token-level recovery declines, while semantic information remains relatively well preserved. We also find that scaling effects are less pronounced in inversion tasks. Finally, our framework demonstrates robust generalization to out-of-distribution clinical data.