We Trained and Open-Sourced a Bilingual Semantic Highlighting Model for Production RAG and AI Search

What this post added

Introduces and open-sources a new bilingual semantic highlighting model (zilliz/semantic-highlight-bilingual-v1) specifically trained for production RAG and AI search. This model addresses limitations of existing semantic highlighting solutions, such as small context windows, poor out-of-domain generalization, weaker multilingual performance, and restrictive licensing. The post details the shortcomings of keyword-based highlighting and existing semantic highlighting approaches (OpenSearch semantic highlighter, Provence/X Provence, Open Provence) and explains the design considerations for the new model, emphasizing strong multilingual performance, large context window, robust generalization, high precision, and a permissive license.