Align Then Adapt: Label-Efficient Adapter Learning for Asymmetric Dense Retrieval

Seiji Maekawa; Moin Aminnaseri; Pouya Pezeshkpour; Estevam Hruschka

cs.IRApr 3, 2026

Align Then Adapt: Label-Efficient Adapter Learning for Asymmetric Dense Retrieval

Authors: Seiji Maekawa, Moin Aminnaseri, Pouya Pezeshkpour, Estevam Hruschka

Organizations: Megagon Labs

Abstract

Dense retrieval systems increasingly face an asymmetry between complex instruction-like queries and relatively simple, static document collections. While stronger embedders can better understand such queries, re-embedding large corpora or fine-tuning large models is often impractical. We propose Efficient Retrieval Adapter (ERA), a query-side adapter learning framework for re-index-free retrieval adaptation. ERA first aligns the embedding spaces of a strong query embedder and a lightweight document embedder using unlabeled corpus documents, and then adapts the aligned query representation with a small number of labeled query-document pairs. Across 126 MAIR retrieval tasks from six domains, ERA improves average nDCG@10 by up to 8.2 points in symmetric settings and by more than 12 points in asymmetric settings, while using substantially fewer labels than supervised adapter training. These results show that retrieval systems can benefit from stronger query understanding without updating backbone embedders or rebuilding document indexes.

Figures & tables

Figure 1: ERA enables re-index-free asymmetric dense retrieval by aligning a strong query embedder with a lightweight document index. Unlike conventional symmetric retrieval, ERA trains a lightweight query-side adapter in two stages: self-supervised alignment using unlabeled corpus documents and supervised adaptation using limited labeled query-document pairs.

	No Parameter	$<1000$	Asymmetric	Domain	Adaptation
Methods	Access	Label Train	Embeddings	Generality	Signal
Zero-shot Retrieval	✓	✗	✗	✓	✗
Embedding Adapter ( Sanjeev and Troynikov, 2024 )	✓	✗	✗	✗	Relevance labels
Search Adaptor ( Yoon et al., 2024 )	✓	✗	✗	✗	Relevance labels
Drift-Adapter ( Vejendla, 2025 )	✓	✗	✓	✓	Alignment Only
ERA (Ours)	✓	✓	✓	✓	Alignment + Relevance

Table 1: Comparison of the proposal, ERA, with existing methods. “Adaptation Signal” summarizes the supervision used to train the adapter. ERA combines cross-embedder alignment with relevance-aware supervised adaptation.

Figure 2: The overview of Efficient Retrieval Adapter. In the alignment stage, the same unlabeled document is encoded by both embedders and used as a positive pair. In the adaptation stage, labeled query-document pairs and hard negatives train the same adapter for retrieval.

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	34.03	34.81	34.85	52.44	37.01	25.25	33.28
	Embedding Adapter	26.18	21.66	30.44	42.09	25.24	18.76	26.55
	Sym. ERA w/o adapt.	34.17	35.01	33.79	53.58	38.41	25.54	33.35
	Sym. ERA	42.14	40.28	40.25	60.77	41.78	32.35	44.08
	Asym. ERA w/o adapt.	43.69	39.15	40.16	58.66	40.65	34.49	47.84
	Asym. ERA	46.59	42.50	46.19	59.27	45.20	38.24	49.28

Table 2: nDCG@10 (%) on MAIR using 20% labeled query-document pairs. Bold indicates the best result per model and domain. Embedding Adapter and Symmetric ERA use the same model for both query and document embedders. Asymmetric ERA uses Qwen3-8B as the query embedder and the base model as the document embedder.

Panel of Figure 3 — Figure 3: nDCG@10 of zero-shot, ERA w/o adaptation, Embedding Adapter, and ERA retrieval at varying train ratios, where we use Qwen3-8B as the query embedder.

Panel of Figure 4 — Figure 4: Domain generality analysis using Qwen3-8B as the query embedder and Qwen3-0.6B as the document embedder, trained on 20% of labeled query-document pairs.

Figure 5: Ablation on negative sampling strategy with varying train ratios.

# Negatives	5	10	50	100	200
nDCG@10 (%)	53.02	52.80	52.89	52.65	52.66

Table 3: Number of negative samples ablation. We set the train ratio to 10%.

Method	nDCG@10	Recall@100	MAP@100
MLP ERA w/o adapt.	50.33	70.79	39.74
MLP ERA	51.84	72.49	41.27
Linear ERA	53.02	72.96	42.39

Table 4: Adapter complexity analysis (%). We set the train ratio to 10%, a query encoder to Qwen3-8B, and a document encoder to Qwen3-0.6B.

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Size	Context	Dimension	HuggingFace / API	License
Qwen3-8B ( Zhang et al., 2025 )	8B	32k	4096	Qwen/Qwen3-Embedding-8B	Apache license 2.0
Qwen3-0.6B ( Zhang et al., 2025 )	0.6B	32k	1024	Qwen/Qwen3-Embedding-0.6B	Apache license 2.0
OpenAI-small ( OpenAI, 2024 )	-	8k	1536	text-embedding-3-small	OpenAI Service Terms 1 1 1 https://openai.com/policies/services-agreement/ [Accessed: May 22, 2026]
BGE-M3 ( Chen et al., 2024 )	0.6B	8k	1024	BAAI/bge-m3	MIT

Appendix

Table 5: Models used in experiments. Dimensions denote the output embedding sizes. Model sizes are not publicly disclosed (-).

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	34.03	34.81	34.85	52.44	37.01	25.25	33.28
	Emb. Adapter	7.48	6.59	14.36	8.92	6.65	4.04	6.60
	Sym. ERA w/o adapt.	34.17	35.01	33.79	53.58	38.41	25.54	33.35
	Sym. ERA	39.80	38.75	39.62	58.45	39.60	29.65	41.02
	Asym. ERA w/o adapt.	43.69	39.15	40.16	58.66	40.65	34.49	47.84
	Asym. ERA	45.04	41.25	42.48	58.47	43.31	35.72	48.65

Appendix

Table 6: nDCG@10 (%) on the MAIR benchmark with 5% of labeled query-document pairs.

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	34.03	34.81	34.85	52.44	37.01	25.25	33.28
	Embedding Adapter	17.37	14.68	22.16	27.52	15.27	11.67	17.51
	Sym. ERA w/o adapt.	34.17	35.01	33.79	53.58	38.41	25.54	33.35
	Sym. ERA	40.77	39.57	38.94	59.36	40.47	31.15	42.43
	Asym. ERA w/o adapt.	43.69	39.15	40.16	58.66	40.65	34.49	47.84
	Asym. ERA	45.70	42.01	44.20	59.33	44.13	37.16	48.59

Appendix

Table 7: nDCG@10 (%) on the MAIR benchmark with 10% of labeled query-document pairs.

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	34.03	34.81	34.85	52.44	37.01	25.25	33.28
	Embedding Adapter	33.24	28.17	37.72	48.70	33.55	24.28	34.05
	Sym. ERA w/o adapt.	34.17	35.01	33.79	53.58	38.41	25.54	33.35
	Sym. ERA	43.65	41.18	43.27	61.57	43.39	34.46	45.15
	Asym. ERA w/o adapt.	43.69	39.15	40.16	58.66	40.65	34.49	47.84
	Asym. ERA	47.27	42.91	47.89	60.50	48.03	38.05	49.49

Appendix

Table 8: nDCG@10 (%) on the MAIR benchmark with 40% of labeled query-document pairs.

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	55.50	69.34	68.05	84.97	58.20	20.99	54.44
	Embedding Adapter	51.49	63.09	65.85	75.59	56.98	19.53	49.82
	Sym. ERA w/o adapt.	55.28	70.35	66.55	84.28	58.43	20.76	54.27
	Sym. ERA	64.17	78.53	73.59	89.03	66.86	26.56	65.78
	Asym. ERA w/o adapt.	65.91	76.80	79.19	89.52	65.44	26.14	68.86
	Asym. ERA	68.62	80.58	81.22	90.20	69.77	30.65	70.81

Appendix

Table 9: Recall@100 (%) on the MAIR benchmark with 20% of labeled query-document pairs.

Base model	Variant	Avg	Academic	Code	Finance	Legal	Medical	Web
BGE-M3	Zero-shot	26.93	30.23	29.99	47.80	30.66	11.83	26.39
	Embedding Adapter	19.31	17.34	25.66	37.17	21.29	7.43	18.91
	Sym. ERA w/o adapt.	27.01	30.49	29.40	49.03	30.22	11.70	26.66
	Sym. ERA	33.29	35.28	35.05	56.08	34.70	14.60	35.03
	Asym. ERA w/o adapt.	33.34	33.43	34.43	53.55	30.05	13.44	37.63
	Asym. ERA	36.20	36.70	40.34	54.55	33.94	16.20	39.44

Appendix

Table 10: MAP@100 (%) on the MAIR benchmark with 20% of labeled query-document pairs.

Panel of Figure 6 — Figure 6: nDCG@10 comparison between zero-shot, ERA trained on all domains, and ERA trained on all domains except the target domain with varying train ratios, where we use Qwen3-8B as the query embedder and Qwen3-0.6B as the document embedder.

Panel of Figure 7 — Figure 7: nDCG@10 difference between ERA trained on all domains and ERA trained on the target domain with varying train ratios, where we use Qwen3-8B as the query embedder and Qwen3-0.6B as the document embedder.

WD $\setminus$ LR	$10^{-5}$	$5\times 10^{-5}$	$10^{-4}$	$5\times 10^{-4}$	$10^{-3}$
$10^{-2}$	52.83	52.59	52.60	52.58	52.60
$10^{-3}$	52.83	52.59	52.57	52.57	52.58
$10^{-4}$	52.84	52.59	52.56	52.57	52.60

Appendix

Table 11: Hyperparameter search results (nDCG@10 %). Rows: weight decay (WD), columns: learning rate (LR). Bold = best configuration.

Method	nDCG@10	Recall@100	MAP@100
Zero-shot	26.30	72.83	22.29
ERA (0.6B $\rightarrow$ 0.6B)	32.95	82.97	28.16
ERA (8B $\rightarrow$ 0.6B)	33.52	82.91	28.74

Appendix

Table 12: Retrieval performance (%) on MS MARCO dev (8.8M passages, exact cosine top-100 search). ERA is trained on MAIR and applied to MS MARCO without any in-domain training.

Method	nDCG@10	R@100	MAP@100
Procrustes w/o adapt.	51.13	70.43	40.59
Sym. ERA w/o adapt.	51.12	70.44	40.59
Procrustes label training	51.58	71.72	41.25
Sym. ERA	51.62	71.75	41.29

Appendix

Table 13: Comparison of the closed-form alignment baseline. The train ratio for labal training is set to 10%.

Base model	Variant	Before	Reranked
BGE-M3	Zero-shot	34.14	53.49
	Embedding Adapter	26.18	50.04
	Sym. ERA w/o adapt.	34.17	53.49
	Sym. ERA	42.14	59.29
	Asym. ERA w/o adapt.	43.68	60.12
	Asym. ERA	46.59	61.86

Appendix

Table 14: nDCG@10 (%) on the MAIR benchmark before and after reranking with Qwen3-Reranker-8B. Top-100 first-stage candidates are reranked. Bold indicates the best result for each base model and column. Asymmetric ERA uses Qwen3-8B as the query encoder and the base model as the document encoder.

Dense Retrieval

Dense retrieval uses learned vector representations to find relevant documents, passages, products, or tools by semantic similarity.

See all 148 papers

Navigation-Informed Embeddings: Dense-Retriever Adaptation from Agent Search Traces
Aug 16, 2026Shrey Shah, Levent OzgurDense Retrieval Agentic Retrieval
Towards Dynamic Dense Retrieval with Routing Strategy
Feb 26, 2026Zhan Su, Fengran Mo, Jinghan Zhang +4Prefix Tuning Dense Retrieval
Université de Montréal, Montréal, Québec, Canada · Clemson University, Clemson, South Carolina, USA · University of Washington, America
OPERA: Online Data Pruning for Efficient Retrieval Model Adaptation
Mar 17, 2026Haoyang Fang, Shuai Zhang, Yifei Ma +5Dataset Pruning Dense Retrieval
Amazon Web Services

Aug 16, 2026J/K move · Enter open · S save

Navigation-Informed Embeddings: Dense-Retriever Adaptation from Agent Search Traces

Shrey Shah, Levent Ozgur

Dense Retrieval Agentic Retrieval Contrastive Learning

Open paper AskPDF arXiv

Agentic retrieval workflows produce query, retrieval, and stopping traces as a byproduct of answering questions. We study how these traces can adapt a deployed dense retriever to changing workflow distributions without new relevance labels, synthetic queries, or LLM judgments. We introduce Navigation-Informed Embeddings (NIE), a family of trace-derived objectives. NIE-Stop turns the stopping document into a soft positive; NIE-Path additionally uses preceding path documents as hard comparisons and imposes ordinal constraints with geometric decay. A BGE encoder adapted from retained source trajectories improves support Recall@20 on an independent target benchmark from 72.2 to 78.0 overall. NIE-Stop reaches 76.9 overall and 52.3 on long paths; NIE-Path raises long-path performance to 55.4, compared with 46.7 for the unadapted encoder. A shuffled-order control under the full path objective loses 3.2 points. Without public-benchmark training, the same adapter also improves nDCG@10 by 1.9 points on standard BEIR HotpotQA. NIE therefore provides a lightweight adaptation channel for settings where trajectories are already retained, with zero incremental labeling cost.

Abstract

Dense Retrieval

Explore similar work

Navigation-Informed Embeddings: Dense-Retriever Adaptation from Agent Search Traces

Towards Dynamic Dense Retrieval with Routing Strategy

OPERA: Online Data Pruning for Efficient Retrieval Model Adaptation