The community members are discussing the challenges of using open-source large language models (LLMs) for chatting with SQL. One community member has tried many models, but found that while the models can generate correct SQL queries, the responses to the main queries are often strange. They believe the natural language to SQL (NLSQL) query is not powerful enough, as even GPT-3.5 struggles to answer all the questions. The community members suggest trying larger models, such as Mixtral, and one mentions that among the models they've used, Zephyr was the best, but still not good enough.
Thanks for your response I have tried too many models. The issue is sometimes models are generating correct sql query, however , response the main query so weird!
I believe the NLSQL query is not powerful enough! Because gpt3.5 is also can't answer all the questions. I wanna see if there's any better options such as using agents. The problem is even agents are tailored for OpenAI models🥴
