aiqbench analysis
This paper introduces LACONIC, a family of sparse retrieval models based on the Llama3 architecture designed to overcome the high memory costs associated with dense vector search. By employing a two-phase training curriculum involving weakly supervised pre-finetuning and hard negative mining, these models achieve state-of-the-art performance while running efficiently on commodity CPUs with significantly reduced index storage requirements. The work represents a significant advancement in making large-scale information retrieval more accessible through learned sparse methods rather than relying solely on GPU-accelerated dense embeddings.
LACONIC: Dense-Level Effectiveness for Scalable Sparse Retrieval via a Two-Phase Training Curriculum
arXiv:2601.01684v2 Announce Type: replace-cross Abstract: While dense retrieval models have been the standard for state-of-the-art information retrieval, their deployment is often constrained by high memory requirements and reliance on GPU accelerators for vector similarity search at scale. Learned sparse retrieval offers a compelling alternative by enabling efficient search via inverted indices, yet it has historically received less attention than dense approaches. In this paper, we introduce LACONIC, a family of learned sparse retrievers based on the Llama3 architecture (1B, 3B, and 8B). We propose a streamlined two-phase training curriculum consisting of (1) weakly supervised pre-finetuning to adapt causal LLMs for bidirectional contextualization and (2) high-signal finetuning using curated hard negatives. Our results demonstrate that LACONIC effectively bridges the performance gap with dense models: the 8B variant achieves a state-of-the-art 60.2 nDCG@10 on the MTEB Retrieval benchmark, ranking 15th on the leaderboard as of February 5th, 2026, while utilizing 74% less index memory than an equivalent dense model. By delivering high retrieval effectiveness on commodity CPU hardware with a fraction of the compute budget required by competing models, LACONIC provides a scalable and efficient solution for real-world search applications. We fully open source our code implementation and trained checkpoints to facilitate reproducibility.
Abstract:While dense retrieval models have been the standard for state-of-the-art information retrieval, their deployment is often constrained by high memory requirements and reliance on GPU accelerators for vector similarity search at scale. Learned sparse retrieval offers a compelling alternative by enabling efficient search via inverted indices, yet it has historically received less attention than dense approaches. In this paper, we introduce LACONIC, a family of learned sparse retrievers based on the Llama3 architecture (1B, 3B, and 8B). We propose a streamlined two-phase training curriculum consisting of (1) weakly supervised pre-finetuning to adapt causal LLMs for bidirectional contextualization and (2) high-signal finetuning using curated hard negatives. Our results demonstrate that LACONIC effectively bridges the performance gap with dense models: the 8B variant achieves a state-of-the-art 60.2 nDCG@10 on the MTEB Retrieval benchmark, ranking 15th on the leaderboard as of February 5th, 2026, while utilizing 74% less index memory than an equivalent dense model. By delivering high retrieval effectiveness on commodity CPU hardware with a fraction of the compute budget required by competing models, LACONIC provides a scalable and efficient solution for real-world search applications. We fully open source our code implementation and trained checkpoints to facilitate reproducibility.
Information Retrieval (cs.IR); Computation and Language (cs.CL)
arXiv:2601.01684 [cs.IR]
(or arXiv:2601.01684v2 [cs.IR] for this version)
https://doi.org/10.48550/arXiv.2601.01684 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 292 words
Related coverage
Inside NVIDIA Rubin GPU Architecture: Powering the Era of Agentic AI
NVIDIA · Jul 21, 2026
Article: Your Next DSL Author is a Language Model
InfoQ · Sep 16, 2026
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Hugging Face · Aug 18, 2026
Introducing Instella-MoE: A State-of-the-Art Fully Open Mixture-of-Experts Language Model
AMD · Jul 24, 2026