hLLM: Single Pass Decoding for Generative Reranking

Abstract:Large language models (LLMs) achieve state-of-the-art generative ranking quality, but the ranking they produce must be decoded, and autoregressive decoding spends one sequential forward pass per emitted token. We observe that the only tokens a ranker must emit are the $N$ ordinal values naming the items in ranked order, and that this narrow, permutation-structured output format admits decoding strategies which are much more efficient than left-to-right generation. We introduce hLLM (Hungarian LLM), a format-specialized decoding strategy that decodes all $N$ ordinals in $O(1)$ forward passes. hLLM reads an $N \times K$ item-position score matrix off the LLM's prefill hidden states with a lightweight self-attention head, then decodes the ordinals as the optimal bipartite assignment of that matrix via the Hungarian algorithm, yielding a valid permutation by construction rather than by repair. Through a systematic study of training signals and backbone adaptation, we show that LoRA-based fine-tuning combined with teacher ranking distillation reaches 28 ms end-to-end inference, a speed-up of $64\times$ while maintaining ranking quality on par with the teacher. We provide a complete ablation decomposing the contributions of architecture, training signal, and backbone adaptation. Our framework connects generative ranking to combinatorial optimization, opening a path toward other $O(1)$-decode mechanisms for real-time ranking.

Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)

arXiv:2609.01807 [cs.LG]

(or arXiv:2609.01807v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2609.01807 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

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.