CLJul 27, 2025

Length Representations in Large Language Models

arXiv:2507.20398v22 citationsh-index: 14EMNLP
Originality Incremental advance
AI Analysis

This provides insights into the internal mechanisms of LLMs for controlling output length, which is incremental but useful for researchers and practitioners in natural language processing.

The study tackled the problem of understanding how large language models internally encode and control output sequence length, finding that multi-head attention mechanisms are critical and that specific hidden units can adjust length without losing semantic informativeness.

Large language models (LLMs) have shown remarkable capabilities across various tasks, that are learned from massive amounts of text-based data. Although LLMs can control output sequence length, particularly in instruction-based settings, the internal mechanisms behind this control have been unexplored yet. In this study, we provide empirical evidence on how output sequence length information is encoded within the internal representations in LLMs. In particular, our findings show that multi-head attention mechanisms are critical in determining output sequence length, which can be adjusted in a disentangled manner. By scaling specific hidden units within the model, we can control the output sequence length without losing the informativeness of the generated text, thereby indicating that length information is partially disentangled from semantic information. Moreover, some hidden units become increasingly active as prompts become more length-specific, thus reflecting the model's internal awareness of this attribute. Our findings suggest that LLMs have learned robust and adaptable internal mechanisms for controlling output length without any external control.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes