Application to text-2-sql generation via LLMs

[scigeek72]

Question

I have been working with text-to-sql generation for the last few months. Very recently, I came to know about inseq module which seems to me to be very interesting (and something most people don't think about in the euphoria of LLMs). It seems to me that, if I could use inseq, this tool can help me understand the relation between a natural language query and the generated SQL statement; for example, what words/tokens in the query lead up to a what part of the SQL statement and so on. Although, I am not sure if it would make sense. I am not a researcher in this area and hence I am not very familiar with all the attempts that have been made over the years.

Is it possible (particularly with models type Qwen 2.5 Coder )?

Also, inseq is not compatible with python 3.13 yet. It would be nice if you guys can fix it.

Thanks