LaSER: Internalizing Explicit Reasoning into Latent Space for Dense Retrieval

TL;DR

LaSER introduces a self-distillation framework that internalizes explicit reasoning into dense retrievers' latent space, enabling reasoning capabilities without the latency of explicit CoT generation, and significantly improves performance on reasoning benchmarks.

Contribution

The paper presents LaSER, a novel training method that internalizes explicit reasoning into the latent space of dense retrievers, bridging explicit reasoning paths with implicit latent thinking.

Findings

Outperforms state-of-the-art baselines on reasoning benchmarks.

Demonstrates robustness across different backbones and model scales.

Effectively combines reasoning depth with inference efficiency.

Abstract

LLMs have fundamentally transformed dense retrieval, upgrading backbones from discriminative encoders to generative architectures. However, a critical disconnect remains: while LLMs possess strong reasoning capabilities, current retrievers predominantly utilize them as static encoders, leaving their potential for complex reasoning unexplored. To address this, existing approaches typically adopt rewrite-then-retrieve pipelines to generate explicit CoT rationales before retrieval. However, this incurs prohibitive latency. In this paper, we propose LaSER, a novel self-distillation framework that internalizes explicit reasoning into the latent space of dense retrievers. Operating on a shared LLM backbone, LaSER introduces a dual-view training mechanism: an Explicit view that explicitly encodes ground-truth reasoning paths, and a Latent view that performs implicit latent thinking. To bridge the gap between these views, we design a multi-grained alignment strategy. Beyond standard output alignment, we introduce a trajectory alignment mechanism that synchronizes the intermediate latent states of the latent path with the semantic progression of the explicit reasoning segments. This allows the retriever to think silently and effectively without autoregressive text generation. Extensive experiments on both in-domain and out-of-domain reasoning-intensive benchmarks demonstrate that LaSER significantly outperforms state-of-the-art baselines. Furthermore, analyses across diverse backbones and model scales validate the robustness of our approach, confirming that our unified learning framework is essential for eliciting effective latent thinking. Our method successfully combines the reasoning depth of explicit CoT pipelines with the inference efficiency of standard dense retrievers.