dosubot[bot]
commented
11 months ago 🤖
Hello,
The issue you're experiencing seems to be related to the conversion of natural language queries to SQL. In the LlamaIndex codebase, the NLStructStoreQueryEngine class is responsible for this conversion. It uses the llm_predictor to convert the natural language query into SQL, and then attempts to run this SQL query. If the conversion from natural language to SQL is not accurate, it could result in invalid SQL being executed, leading to an OperationalError.
As for the warnings about not being able to load the OpenAI model, it's likely because the llm_predictor object, which is used to convert natural language queries to SQL and to synthesize responses, is not able to load the required model. This could be due to a variety of reasons such as the model file not being found, insufficient resources to load the model, or the model being incompatible with the version of the library being used.
There are two similar issues that have been solved in the LlamaIndex repository:
-
Bug: text-to-SQL example fails with the new LlamaCPP fallback which was closed by a pull request: Improve SQL query parsing
-
Bug: sqlite3.Warning: You can only execute one statement at a time in NLSQLTableQueryEngine The suggested solution is to update to the latest version of llama-index. The issue has been addressed in the latest codebase. Relevant PR: https://github.com/jerryjliu/llama_index/pull/6782
As for the SQLStructStoreIndex.from_documents method, I wasn't able to find specific documentation within the repository. However, based on the method name and typical usage in similar frameworks, it's likely that this method is used to create an instance of the SQLStructStoreIndex class from a list of documents, where each document is a structured data item that can be indexed and queried.
I hope this helps! If you have any other questions or need further clarification, feel free to ask.
Sources
- [Bug]: text-to-SQL example fails with the new LlamaCPP fallback
- [Bug]: sqlite3.Warning: You can only execute one statement at a time in NLSQLTableQueryEngine
- llama_index/indices/struct_store/sql_query.py
- benchmarks/struct_indices/spider/generate_sql.py
- docs/examples/index_structs/struct_indices/SQLIndexDemo.ipynb
This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.
Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.
If you want to continue the conversation, start your reply with @dosu-bot.
vp999
devstein
yonitjio
Bug Description
I was trying to use SQLStructStoreIndex as query engine to interact with sqllite database. The code is erroring out because the database sql execution code is recieving the natural langauge text in addition to sql text.
Version
0.8.5.post1
Steps to Reproduce
do not set open api token , this way it uses llama2 as local default
sql_database = SQLDatabase.from_uri("sqlite:///database/chinook.db")
index = SQLStructStoreIndex.from_documents( [], sql_database=sql_database,) query_engine = index.as_query_engine()
query_engine.query("who are the top selling artists?")
output error : OperationalError: (sqlite3.OperationalError) near "Question": syntax error [SQL: Question: Who are the top selling artists?
SQLQuery: SELECT ArtistId, Name, SUM(UnitPrice * Quantity) AS TotalSales FROM invoices JOIN invoice_items ON invoices.InvoiceId = invoice_items.InvoiceId JOIN tracks ON invoice_items.TrackId = tracks.TrackId JOIN artists ON tracks.ArtistId = artists.ArtistId GROUP BY ArtistId, Name ORDER BY TotalSales DESC;] (Background on this error at: https://sqlalche.me/e/20/e3q8)
Relevant Logs/Tracbacks