tenfingers/tsql-docs-rlat

T-SQL文档 — rlat知识模型是一个基于微软SQL文档（特别是T-SQL文档部分）预构建的知识模型，专为检索增强的编码助手和大型语言模型（LLM）上下文注入而设计。该数据集面向编写T-SQL查询、存储过程和函数的SQL Server、Azure SQL和Fabric Warehouse开发人员。数据集包含两个版本：一个版本在首次查询时从GitHub获取源文本并本地缓存（远程版本），另一个版本是完全自包含的离线版本（捆绑版本）。数据集使用Resonance Lattice框架构建，提供了编码器、压缩和构建过程的具体技术细节。许可信息涵盖了源内容和知识模型结构，分别适用不同的许可协议。数据集旨在帮助开发人员更高效地编写T-SQL代码，并提供了示例查询和使用说明。

The T-SQL Docs — rlat knowledge model is a prebuilt knowledge model based on Microsofts SQL documentation, specifically scoped to T-SQL (Transact-SQL) documentation. It is designed for retrieval-augmented coding assistants and LLM context injection, targeting SQL Server, Azure SQL, and Fabric Warehouse developers writing T-SQL queries, stored procedures, and functions. The dataset includes two variants: a remote version that fetches source text from GitHub on first query and caches locally, and a bundled version that is fully self-contained for offline use. Built using the Resonance Lattice framework, it details the encoder (BGE-large-en-v1.5 with random projection heads), compression (8-bit quantized registry, zstd compressed), and build process. Licensing covers both the source content (CC BY 4.0 for documentation, MIT for code samples) and the artifact structure (Business Source License 1.1). The dataset facilitates efficient T-SQL coding with sample queries and usage instructions provided.