MACA: A Framework for Distilling Trustworthy LLMs into Efficient Retrievers

TL;DR

MACA introduces a metadata-aware distillation framework that transforms a large language model re-ranker into an efficient, metadata-conditioned retriever, significantly improving accuracy while avoiding costly online LLM calls.

Contribution

The paper presents MACA, a novel framework for distilling trustworthy LLM re-rankers into compact retrievers using metadata-aware prompts and a new training objective, enhancing retrieval accuracy and efficiency.

Findings

MACA surpasses baseline accuracy by 5-3 points on proprietary and public datasets.

Students outperform pretrained encoders, e.g., MiniLM accuracy doubles from 0.23 to 0.48.

MACA enables retrieval-augmented generation without online LLM calls.

Abstract

Modern enterprise retrieval systems must handle short, underspecified queries such as ``foreign transaction fee refund'' and ``recent check status''. In these cases, semantic nuance and metadata matter but per-query large language model (LLM) re-ranking and manual labeling are costly. We present Metadata-Aware Cross-Model Alignment (MACA), which distills a calibrated metadata aware LLM re-ranker into a compact student retriever, avoiding online LLM calls. A metadata-aware prompt verifies the teacher's trustworthiness by checking consistency under permutations and robustness to paraphrases, then supplies listwise scores, hard negatives, and calibrated relevance margins. The student trains with MACA's MetaFusion objective, which combines a metadata conditioned ranking loss with a cross model margin loss so it learns to push the correct answer above semantically similar candidates with mismatched topic, sub-topic, or entity. On a proprietary consumer banking FAQ corpus and BankFAQs, the MACA teacher surpasses a MAFA baseline at Accuracy@1 by five points on the proprietary set and three points on BankFAQs. MACA students substantially outperform pretrained encoders; e.g., on the proprietary corpus MiniLM Accuracy@1 improves from 0.23 to 0.48, while keeping inference free of LLM calls and supporting retrieval-augmented generation.