Progress and Tradeoffs in Neural Language Models
This highlights a critical performance-efficiency problem for deploying neural language models on resource-constrained devices like mobile platforms, though it is incremental in nature.
The paper examines the tradeoff between quality and performance in neural language models, finding that on a Raspberry Pi, large increases in latency and energy usage correspond to minimal improvements in perplexity, while this effect is less pronounced on a desktop.
In recent years, we have witnessed a dramatic shift towards techniques driven by neural networks for a variety of NLP tasks. Undoubtedly, neural language models (NLMs) have reduced perplexity by impressive amounts. This progress, however, comes at a substantial cost in performance, in terms of inference latency and energy consumption, which is particularly of concern in deployments on mobile devices. This paper, which examines the quality-performance tradeoff of various language modeling techniques, represents to our knowledge the first to make this observation. We compare state-of-the-art NLMs with "classic" Kneser-Ney (KN) LMs in terms of energy usage, latency, perplexity, and prediction accuracy using two standard benchmarks. On a Raspberry Pi, we find that orders of increase in latency and energy usage correspond to less change in perplexity, while the difference is much less pronounced on a desktop.