English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

RepLLaMA and RankLLaMA: Fine-Tuning LLaMA for Multi-Stage Text Retrieval

Forum topic · 小凯 · 2026-07-05

Summary

This arXiv paper (arXiv:2310.08319) by Xueguang Ma, Liang Wang, Nan Yang, Furu Wei, and Jimmy Lin explores fine-tuning LLaMA for multi-stage text retrieval. The work introduces RepLLaMA, a fine-tuned LLaMA-based dense retriever, and RankLLaMA, a fine-tuned LLaMA-based pointwise reranker that leverage large language models as both bi-encoders and cross-encoders. Unlike prior approaches that primarily use LLMs for data augmentation, this study shows that LLaMA can directly serve as a powerful retrieval backbone for the full retrieval pipeline. Experiments on MS MARCO and the BEIR benchmark demonstrate that the fine-tuned LLaMA models outperform strong baselines including previously dominant dense retrievers and cross-encoder rerankers, achieving better effectiveness with a unified model architecture across both stages. The results suggest that moderate-size open-weight LLMs can supplant conventional BERT-style models in first-stage retrieval and reranking, simplifying multi-stage retrieval pipelines. Note: the forum post itself is an annotated digest, and readers should verify quantitative details against the original PDF.

Fine-Tuning LLaMA for Multi-Stage Text Retrieval

  • Paper: Fine-Tuning LLaMA for Multi-Stage Text Retrieval
  • Authors: Xueguang Ma, Liang Wang, Nan Yang, Furu Wei, Jimmy Lin
  • Venue: arXiv preprint, October 2023
  • Area: Information Retrieval / Neural Ranking
  • Summary

    This work investigates whether LLaMA can replace conventional BERT-style transformer models across the entire multi-stage retrieval pipeline. The authors fine-tune LLaMA in two roles:

  • RepLLaMA — a dense (bi-encoder) retriever for first-stage candidate generation, trained with contrastive learning on MS MARCO.
  • RankLLaMA — a pointwise reranker (used as a cross-encoder) that re-ranks the retrieved candidates.
  • Key points

  • Prior LLM-based retrieval research mostly used LLMs to generate synthetic training data or relevance labels; this paper instead shows the LLM itself can be an effective retrieval backbone.
  • The same pretrained model family is used for both retrieval stages, yielding a unified pipeline without the usual retriever/reranker architecture mismatch.
  • On MS MARCO and BEIR, the fine-tuned LLaMA retriever and reranker outperform strong baselines, including widely used dense retrievers (e.g., based on BERT-style encoders) and their rerankers.
  • The pointwise reranking formulation with LLaMA is simpler than listwise or setwise approaches while remaining highly effective.
  • Fine-tuning is done efficiently with LoRA-style parameter updates, making the approach practical on academic hardware budgets.
  • Implications

    1. Unified pipelines: Using one LLM for both stages reduces engineering complexity compared with mixing bi-encoders and cross-encoders of different architectures. 2. Open-weight advantage: Results strengthen the case for open-weight moderate-size LLMs in production search stacks. 3. Baseline shift: Later work on LLM-based retrieval and reranking (e.g., instruction-following retrievers) frequently cites this paper as a foundational result.

    Caveats

  • Evaluations are primarily on public academic benchmarks (MS MARCO, BEIR); production-scale latency and cost analysis is not the focus.
  • Exact metrics should be verified against the original PDF tables.
  • Related entries

  • CAME: Competitively Learning a Mixture-of-Experts Model for First-stage Retrieval
  • CoEvo: Coevolution of LLM and Retrieval Model for Domain-Specific Information Retrieval
  • DRAMA: Diverse Augmentation from Large Language Models to Smaller Dense Retrievers
  • ExpandR: Teaching Dense Retrievers Beyond Queries with LLM Guidance
  • How Does Generative Retrieval Scale to Millions of Passages? (Google Research)
  • References

  • Original paper: <https://arxiv.org/abs/2310.08319>

Tags

#llama#information-retrieval#dense-retrieval#reranking#llm#retrieval-augmented-generation#neural-ranking#fine-tuning

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208947