Enhancing Latent Computation in Transformers with Latent Tokens
This work addresses the adaptability of large language models, offering a lightweight, parameter-efficient method for improving performance, though it is incremental as it builds on existing token augmentation strategies.
The authors tackled the problem of enhancing large language model performance by introducing latent tokens, which are auxiliary dummy tokens that steer Transformer decoding via attention, and found that their method noticeably outperforms baselines, especially in out-of-distribution generalization scenarios.
Augmenting large language models (LLMs) with auxiliary tokens has emerged as a promising strategy for enhancing model performance. In this work, we introduce a lightweight method termed latent tokens; these are dummy tokens that may be non-interpretable in natural language but steer the autoregressive decoding process of a Transformer-based LLM via the attention mechanism. The proposed latent tokens can be seamlessly integrated with a pre-trained Transformer, trained in a parameter-efficient manner, and applied flexibly at inference time, while adding minimal complexity overhead to the existing infrastructure of standard Transformers. We propose several hypotheses about the underlying mechanisms of latent tokens and design synthetic tasks accordingly to verify them. Numerical results confirm that the proposed method noticeably outperforms the baselines, particularly in the out-of-distribution generalization scenarios, highlighting its potential in improving the adaptability of LLMs.