Parallel Context-of-Experts Decoding for Retrieval Augmented Generation

TL;DR

The paper introduces Pced, a decoding framework that enables cross-document reasoning in retrieval-augmented generation by treating documents as experts and synchronizing their outputs without shared attention.

Contribution

Pced shifts evidence aggregation to decoding, allowing multi-document reasoning without shared attention, improving speed and interaction in retrieval-augmented generation.

Findings

Recovers cross-document reasoning capabilities

Maintains speed by avoiding shared attention

Uses retrieval-aware contrastive decoding

Abstract

Retrieval Augmented Generation faces a trade-off: concatenating documents in a long prompt enables multi-document reasoning but creates prefill bottlenecks, while encoding document KV caches separately offers speed but breaks cross-document interaction. We propose Parallel Context-of-Experts Decoding (Pced), a training-free framework that shifts evidence aggregation from the attention mechanism to the decoding. Pced treats retrieved documents as isolated "experts", synchronizing their predictions via a novel retrieval-aware contrastive decoding rule that weighs expert logits against the model prior. This approach recovers cross-document reasoning capabilities without constructing a shared attention across documents.