hLLM: Single Pass Decoding for Generative Reranking

Researchers introduced hLLM (Hungarian LLM), a decoding strategy that generates rankings from large language models in a single forward pass, achieving 28 ms end-to-end inference, a 64x speed-up over autoregressive decoding while maintaining ranking quality on par with the teacher model. The method reads an N x K item-position score matrix from the LLM's prefill hidden states and decodes ordinals via the Hungarian algorithm, ensuring a valid permutation by construction.

arXiv:2609.01807v1 Announce Type: new 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.