Text-to-SQL using RLHF + Execution Reward
Convert Natural Language to SQL, strictly validated and safely executed on local SQLite databases.
1. Configuration & Input
Select a Sample Question
The database will be selected automatically.
⚠️ Please select a Database first, then type your custom question below:
Select Database
📋 Database Structure
Use these exact names! Table names are Dark, Column names are Light.
ALBUM ( albumid, title, artistid )
ARTIST ( artistid, name )
CUSTOMER ( customerid, firstname, lastname, company, address, city, state, country, postalcode, phone, fax, email, supportrepid )
EMPLOYEE ( employeeid, lastname, firstname, title, reportsto, birthdate, hiredate, address, city, state, country, postalcode, phone, fax, email )
GENRE ( genreid, name )
INVOICE ( invoiceid, customerid, invoicedate, billingaddress, billingcity, billingstate, billingcountry, billingpostalcode, total )
INVOICELINE ( invoicelineid, invoiceid, trackid, unitprice, quantity )
MEDIATYPE ( mediatypeid, name )
PLAYLIST ( playlistid, name )
PLAYLISTTRACK ( playlistid, trackid )
TRACK ( trackid, name, albumid, mediatypeid, genreid, composer, milliseconds, bytes, unitprice )
2. Execution Results
Query Result Table
1 | 2 | 3 |
|---|---|---|
Diagnostics & Telemetry
(Simulates the heavy RLHF training workload by running hundreds of complex SQL queries concurrently to test SQLite multi-threading performance.)
(Live telemetry tracking the most common SQL failures. Populates automatically when queries fail in the Inference tab.)
Error counts
1 | 2 | 3 |
|---|---|---|
Recent errors
1 | 2 | 3 |
|---|---|---|
Select error type
(Analyzes which specific SQL clauses—SELECT, WHERE, JOIN, etc.—are most prone to errors during natural language generation.)
Success/failure by op
1 | 2 | 3 |
|---|---|---|