Confident Adaptive Language Modeling
This addresses the issue of slow and costly inference for users of large language models, though it is incremental as it builds on early exit decoding methods.
The paper tackles the problem of high computational cost in large language models at inference time by introducing Confident Adaptive Language Modeling (CALM), a framework that dynamically allocates compute per input and generation timestep, achieving up to 3x speedup while maintaining performance.
Recent advances in Transformer-based large language models (LLMs) have led to significant performance improvements across many tasks. These gains come with a drastic increase in the models' size, potentially leading to slow and costly use at inference time. In practice, however, the series of generations made by LLMs is composed of varying levels of difficulty. While certain predictions truly benefit from the models' full capacity, other continuations are more trivial and can be solved with reduced compute. In this work, we introduce Confident Adaptive Language Modeling (CALM), a framework for dynamically allocating different amounts of compute per input and generation timestep. Early exit decoding involves several challenges that we address here, such as: (1) what confidence measure to use; (2) connecting sequence-level constraints to local per-token exit decisions; and (3) attending back to missing hidden representations due to early exits in previous tokens. Through theoretical analysis and empirical experiments on three diverse text generation tasks, we demonstrate the efficacy of our framework in reducing compute -- potential speedup of up to $\times 3$ -- while provably maintaining high performance.