evgenypal/k8s-docs-rag-bench
收藏资源简介:
k8s-docs-rag-bench是一个完全基于Kubernetes官方文档的小型基准数据集,用于检索增强问答(RAG)任务。该数据集包含问答对、语义分块的文档语料库以及LLM评估标签,旨在支持密集/稀疏/混合检索技术测试、基于LoRA的小型开源LLM训练以及论文中管道消融和LoRA排名研究的复现。数据集基于Kubernetes文档的特定快照构建,所有问答对均经过人工检查,并采用页面级分割以确保泛化能力。
k8s-docs-rag-bench is a small-scale benchmark dataset entirely based on the official Kubernetes documentation, designed for retrieval-augmented generation (RAG) tasks. This dataset includes question-answer pairs, semantically chunked document corpora, and LLM evaluation labels, aiming to support tests of dense/sparse/hybrid retrieval techniques, fine-tuning of small open-source LLMs based on LoRA, and reproduction of pipeline ablation and LoRA ranking studies in academic papers. The dataset is constructed from a specific snapshot of Kubernetes documentation, with all question-answer pairs manually reviewed and page-level segmentation adopted to ensure generalization capability.





