is there an option does not call openai

It just calls the normal LLM (text-davinci-003 by default)

However, it's only sending the table schema(s) plus the user queries, so not too data sensitive?

If that's still an issue though, yea you can use a local LLM, but you'll need powerful hardware to run it

Thanks Logan, I was thinking about this model. https://huggingface.co/juierror/text-to-sql-with-table-schema Is there anything in huggingface that you would suggest that would perform well for text-to-sql functionality?

Thinking about local LLM option because openai api has a quota for usage and because of cost.

That might be a good option, especially if you are only using a single table at a time

But to be honest, if all you need is the text-to-sql part, it might be easier to use that model directly, without llama-index in the mix

Thanks Logan, I tried out the huggingface text to sql and it does not work as well as the openai api method. The generated sql queries are wrong. Can you please point me to the place in the llama_index source code where you process the user input and call openai api for text to sql?

Right at this line is where it generates the text-to-sql https://github.com/jerryjliu/llama_index/blob/4c0cf6ccc9f51eac9999e9d03548374468978c5a/llama_index/indices/struct_store/sql_query.py#L163C43-L163C43