reproduce on Spider test #55
Description
Hello, I’m running the code you provided and noticed that it uses the Spider 1.0 dev split by default. I would like to reproduce the 86.2% accuracy on the Spider leaderboard using GPT-4 during testing.
Should the parameters I use be exactly the same as those in the run_dail_sql.sh script?
Are there any additional preprocessing steps or parameter adjustments needed?
Thank you for your great work and for open-sourcing the code. I really appreciate any guidance or corrections you can provide regarding my setup!