How do I see exactly what is being sent to the LLM provider when using NLSQLTableQueryEngine?

[szhydkov]

I am testing a basic text-to-SQL tool, following some basic llamaindex guides. I am using NLSQLTableQueryEngine to interact with the mysql database. However, I struggle to find exactly how the engine gets the schema and what prompt it send to the API provider (I am using Groq) to get the SQL query generated.

In the NLSQLRetriever code I found this:

response_str = self._llm.predict(
            self._text_to_sql_prompt,
            query_str=query_bundle.query_str,
            schema=table_desc_str,
            dialect=self._sql_database.dialect,
        )

however, according the info log, table_desc_str is very verbose and has 50k tokens, way beyond the context window. Hence, I am trying how this is possible. Thanks!