Online Cascade Learning for Efficient Inference over Streams
This addresses the problem of efficient stream processing for applications requiring LLMs, offering a significant cost reduction, though it is incremental as it builds on existing cascade and imitation learning ideas.
The paper tackles the high computational cost of LLM inference for answering queries on data streams by proposing online cascade learning, which uses a cascade of models with a deferral policy to achieve similar accuracy while reducing inference costs by up to 90% across four benchmarks.
Large Language Models (LLMs) have a natural role in answering complex queries about data streams, but the high computational cost of LLM inference makes them infeasible in many such tasks. We propose online cascade learning, the first approach to address this challenge. The objective here is to learn a "cascade" of models, starting with lower-capacity models (such as logistic regression) and ending with a powerful LLM, along with a deferral policy that determines the model to be used on a given input. We formulate the task of learning cascades online as an imitation-learning problem, where smaller models are updated over time imitating the collected LLM demonstrations, and give a no-regret algorithm for the problem. Experimental results across four benchmarks show that our method parallels LLMs in accuracy while cutting down inference costs by as much as 90% with strong robustness against input distribution shifts, underscoring its efficacy and adaptability in stream processing.