Mixtures of In-Context Learners

TL;DR

This paper introduces MoICL, a novel method that combines subsets of demonstrations as experts to improve in-context learning performance, efficiency, and robustness without increasing memory demands.

Contribution

MoICL is a new approach that learns to merge demonstration subsets as experts, enhancing performance, reducing inference time, and increasing robustness compared to standard ICL.

Findings

Improves accuracy on 5 out of 7 classification datasets.

Reduces inference time while maintaining performance.

Increases robustness to out-of-domain, imbalanced, and noisy data.

Abstract

In-context learning (ICL) adapts LLMs by providing demonstrations without fine-tuning the model parameters; however, it does not differentiate between demonstrations and quadratically increases the complexity of Transformer LLMs, exhausting the memory. As a solution, we propose Mixtures of In-Context Learners (MoICL), a novel approach to treat subsets of demonstrations as experts and learn a weighting function to merge their output distributions based on a training set. In our experiments, we show performance improvements on 5 out of 7 classification datasets compared to a set of strong baselines (up to +13\% compared to ICL and LENS). Moreover, we enhance the Pareto frontier of ICL by reducing the inference time needed to achieve the same performance with fewer demonstrations. Finally, MoICL is more robust to out-of-domain (up to +11\%), imbalanced (up to +49\%), or noisy demonstrations (up to +38\%) or can filter these out from datasets. Overall, MoICL is a more expressive approach to learning from demonstrations without exhausting the context window or memory.