ggerganov/llama.cpp Released b11818 with Server Rerank Updates
Llama.cpp release b11818 introduces typed content input and various multimodal handling improvements for the server's rerank endpoint. These updates include proper validation to reject multimodal rerank inputs when models lack a rerank chat template, along with improved error codes. These enhancements improve the reliability and error-handling robustness of local AI server deployments using multimodal reranking models. Developers building search and retrieval systems benefit from more predictable API behavior and standardized input formats. The release correctly translates media download or base64 decoding failures to HTTP 400 status codes instead of 500, and rejects negative top_n parameters. Additionally, it updates the documentation to note that Text Embeddings Inference (TEI) input accepts the same shape as Jina.
## BACKGROUND
Llama.cpp is an open-source software library that enables efficient inference of Large Language Models on various hardware backends. The rerank endpoint is a specialized server feature used to score and re-order a list of candidate documents against a specific query for retrieval-augmented generation pipelines.