Externally-transfer performance after distillation remains mixed, so the evidence supports compression of teacher rankings under matched retrieval protocols.
Abstract
Large embedding models improve retrieval quality, but serving large encoders online is expensive. We study whether a compact retriever can learn teacher ranking behavior from score vectors without access to teacher hidden states. The student trains on rows built from ground-truth positives and negative candidates produced by our data generation pipeline; we evaluate student-teacher hard-negative mining separately as an extension. We use a row-centered score-vector objective, a memory-efficient implementation of uniform all-pairs PairMSE loss. On a fixed eight-task evaluation panel, our distillation protocol recovers up to 50% of the base-to-teacher gap. The distilled 0.6B student is 4.7 times faster for query encoding and 9.7 times faster for document encoding than sequential online teacher fusion. External-transfer performance after distillation remains mixed, so our evidence supports compression of teacher rankings under matched retrieval protocols.
Large Language Models (LLMs) have emerged as powerful assets for recommender systems. However, deploying them as generative recommenders or zero-shot rankers at web-scale remains bottlenecked by prohibitive computational overhead and grounding challenges. In this paper, we revitalize the classic, highly efficient two-tower retrieval architecture by adapting LLMs as semantic representation backbones rather than generative engines. We introduce an LLM-native two-tower framework engineered for high-throughput, large-scale retrieval. Our architecture introduces several key innovations: a shared LLM encoder for joint user-item modeling, End-Of-Sentence (EOS) token pooling for compact sequence embedding, cross-dataset transfer learning, knowledge distillation from powerful cross-encoder teachers, and latent reasoning within the user tower. Extensive evaluation across three public benchmarks demonstrates that cross-encoder architecture outperforms current state-of-the-art (SoTA) models, while the efficient two-tower student achieves SoTA-comparable retrieval performance. Furthermore, experiments on internal large-scale production systems yield substantial topline retrieval improvements along with high resilience to model staleness and superior data scaling. Our findings demonstrate that when augmented with modern representation learning, the traditional two-tower paradigm remains an exceptionally competitive and practical solution for industrial retrieval systems.
Zhe Xu, Prachi Agrawal, Kavosh Asadi et al.· 0 citations
A unified pipeline deployed at Walmart that addresses both signal quality and model evolution is presented, and a Warm-Start Distillation technique that transfers domain-specific expertise from the legacy model to the new backbone is introduced.
Zhen Yang, Juexin Lin, Hongwei Shang et al.· Annual International ACM SIG...· 1 citation
This work introduces Giga-Embeddings, a family of text embedding models designed to combine strong retrieval quality with efficient serving, and trains the compact model using a dimension-agnostic objective that aligns teacher and student similarity distributions.
Egor Kolodin, Egor Krasnoperov, Evgeniy Kosarev et al.· 0 citations
Visual document retrieval (VDR) is dominated by multi-billion-parameter models that are slow to index at full corpus scale and expensive to serve. Prior compression routes either train a smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector retriever end-to-end. We present DistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwise cosine alignment loss. All supervision comes from the frozen teacher's embedding space, which was itself trained with relevance supervision, so the student objective needs no relevance labels, negative sampling, or contrastive term. We match VDR's text-query and image-document input asymmetry with an asymmetric encoder-only student that concentrates visual capacity on the document side and keeps the query side at 70M parameters. We release two variants that share the same encoders and training and differ only in the document encoder's visual-tile budget: DistilVDR-HiRes attains 61.74 average NDCG@5 on ViDoRe v1+v2+v3 (86.9% of the 8B teacher) and leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark, while DistilVDR-Fast attains 59.98 at a 3 times smaller visual-token budget. Both variants store one million documents in a 15.6 times smaller index than the strongest sub-1B multi-vector baseline and index the corpus an order of magnitude faster. The code is available at https://github.com/Ryenhails/NanoVDR.
Advancements in retrieval models necessitate re-indexing, a computationally expensive process for large-scale production environments. While updating only the query encoder and continuing to use the old index could be a promising middle ground, dense retrieval systems suffer severe performance drops in this setting. We investigate whether Learned Sparse Retrieval (LSR) can mitigate this backward compatibility issue, as its lexical matching may provide a stable term-based anchor to preserve compatibility across model versions. Experiments on BEIR and the streaming settings of LoTTE show that upgrading only the query encoder causes only a small effectiveness drop in LSR when no mitigation applied, whereas dense retrieval fails severely. We explore lightweight query adaptation methods including ranking fusion, representation fusion, and minimal-training adapters to further improve compatibility. These approaches significantly improve backward compatibility on BEIR and effectively reduce performance loss in streaming retrieval. Code: https://github.com/JingfenQiao/LSR-BC.git
Jingfen Qiao, Gabrielle Poerwawinata, Thong Nguyen et al.· Annual International ACM SIG...· 0 citations
This work proposes a distillation approach based on ranking supervision that consistently outperforms supervised fine-tuning as well as FKL and RKL baselines in Python code generation, multilingual generation, and data-science scenarios and offers guidance for future research in model compression.
Zhe Ding, Hui Ji, Su Pan et al.· Neural Networks· 0 citations
