Gatekeeper: Improving Model Cascades Through Confidence Tuning

TL;DR

Gatekeeper introduces a novel loss function to calibrate smaller models in cascades, enabling better task handling and deferral to larger models, thereby improving resource efficiency across diverse tasks.

Contribution

The paper proposes a new loss function called Gatekeeper for calibrating small models in cascades, enhancing deferral accuracy without architectural changes.

Findings

Significant improvement in deferral performance across tasks.

Applicable to various architectures and domains.

Broadly improves resource efficiency in model cascades.

Abstract

Large-scale machine learning models deliver strong performance across a wide range of tasks but come with significant computational and resource constraints. To mitigate these challenges, local smaller models are often deployed alongside larger models, relying on routing and deferral mechanisms to offload complex tasks. However, existing approaches inadequately balance the capabilities of these models, often resulting in unnecessary deferrals or sub-optimal resource usage. In this work we introduce a novel loss function called Gatekeeper for calibrating smaller models in cascade setups. Our approach fine-tunes the smaller model to confidently handle tasks it can perform correctly while deferring complex tasks to the larger model. Moreover, it incorporates a mechanism for managing the trade-off between model performance and deferral accuracy, and is broadly applicable across various tasks and domains without any architectural changes. We evaluate our method on encoder-only, decoder-only, and encoder-decoder architectures. Experiments across image classification, language modeling, and vision-language tasks show that our approach substantially improves deferral performance.