Teaching Nemotron Greek: Mining a Corpus, Adapting Retrieval, and Grounding Generation for Modern Greek across Specialist Domains
让Nemotron学习希腊语:针对专业领域现代希腊语的语料库挖掘、检索适配与生成落地
专题命中 视觉定位与Grounding :grounding(title);分类 cs.AI
AI总结 本研究针对现代希腊语缺失于Nemotron检索模型及主流多语言基准的问题,提出Nemotron检索栈的端到端适配方案,含语料库挖掘等步骤,推出首个希腊语RAG基准HERA,适配后的模型在多项指标上显著提升。
Comments 15 pages, 10 figures, 7 tables. Includes release of the HERA benchmark and Sophea Nemo RAG models