Learning log
Text-to-SQL on Fireworks
The interview task: put a plain-English interface over a database, and pick a model that’s good enough and cheap enough. These are the notes from building it, one chapter at a time.
Where you areBenchmark run and results checked. Next: write 3 harder questions.
- How it worksThe four moving parts, and why the model never sees your data
- The test setWhy you can’t pick a model by trying a few questions
- Results9 models, 23 questions, and a 26× price gap for the same score
- What went wrongThe marker was wrong before the models were
- Your exerciseWrite 3 harder questions that split the models
- GlossaryThe terms, in plain English