Chinese Brief
中文案例导读
Distyl 公开的代码代理与软件工程案例,来源为 官方页面,复核于 2026-06-27T06:31:07Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。
CASE EVIDENCE / A RECORD
Distyl 公开的代码代理与软件工程案例,来源为 官方页面,复核于 2026-06-27T06:31:07Z。
原始记录:Distyl uses fine-tuned GPT-4o for enterprise text-to-SQL generation
Chinese Brief
Distyl 公开的代码代理与软件工程案例,来源为 官方页面,复核于 2026-06-27T06:31:07Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。
任务
这是一个围绕软件工程任务执行的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Distyl applied a fine-tuned GPT-4o model to text-to-SQL workflows including query reformulation, intent classification, chain-of-thought/self-correction, and SQL generation for enterprise data tasks.
公开材料提供原始证据链接和可访问产物,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:OpenAI reports Distyl's fine-tuned GPT-4o reached 71.83% execution accuracy and ranked first on the BIRD-SQL leaderboard at the time of the post.
GPT-4o (Aug) 在该案例中承担软件工程任务执行相关的生成、分析、编排或实现角色。 原始资料写作:Fine-tuned GPT-4o supplied the domain-specific natural-language-to-SQL reasoning and generation capability used to improve execution accuracy across SQL tasks.
当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:The public evidence is an OpenAI partner success paragraph rather than a Distyl implementation deep dive. The page explicitly ties GPT-4o fine-tuning to selecting gpt-4o-2024-08-06 as the base model.