RankLLaMA

Stub. Created as a placeholder — expand with vault-sourced content.

Definition

RankLLaMA is a LLaMA-based pointwise reranker (from the RepLLaMA/RankLLaMA line, Tevatron) — a decoder LLM fine-tuned to score query-document relevance, representing the LLM-reranker end of the cost/quality spectrum.

Vault references (existing coverage)

TODO

  • Cite the source paper; contrast with prompt-based RankGPT (no fine-tuning).