Thread by @RLanceMartin

Text-to-SQL 📒

LLMs unlock a natural language interface with structured data. Part 4 of our initiative to improve @LangChainAI docs shows how to use LLMs to write / execute SQL queries w/ chains and agents. Thanks @manuelsoria_ for work on the docs:
https://t.co/CyOqp5I3TMpython.langchain.com/docs/use_cases…

1/ Text-to-SQL is an excellent LLM use-case: many ppl can describe what they want in natural language, but have difficultly mapping that to a specific SQL queries. LLMs can bridge this gap, e.g., see:
https://t.co/b0NMkHPe9xarxiv.org/pdf/2204.00498…

2/ create_sql_query_chain( ) maps from natural language to a SQL query: pass the question and the database into the chain, and get SQL out. Run the query on the database easily:

3/ The LangSmith trace is a great way to see that the chain employs ideas from the paper above: give LLM a CREATE TABLE description for each table and and three example rows in a SELECT statement. This gives the LLM context about the db structure:
https://t.co/Pqu86RFcJPsmith.langchain.com/public/c8fa52e…

4/ Extending this, SQLDatabaseChain will generate the query, execute if, and also synthesize the result in natural language. This creates a natural language wrapper around a SQL DB w/ input and output:
https://t.co/avT4kRVIiosmith.langchain.com/public/7f202a0…

5/ Finally, SQL agents can be used for more complex tasks (multi-query) and can recover from errors. The trace shows how a ReAct agent can use a toolkit of SQL operations (read table, write query, run query) :
https://t.co/zUrXja5bzVsmith.langchain.com/public/a86dbe1…

6/ For more, see blog post and webinar w/ @fpingham and @JonZLuo:

https://t.co/dtZQY4gAMLblog.langchain.dev/llms-and-sql/

7/ Finally, for more on the community initiative to improve the docs, see Part 3 on extraction:

https://twitter.com/RLanceMartin/status/1687883380631932928?s=20