https://community.databricks.com/t5/databricks-tv/improving-text2sql-performance-with-ease-on-databricks/ba-p/81083 <P><IFRAME src="https://www.youtube.com/embed/tzjqhIn7dvo?si=RRCR8kwA8CLoM6Wp" width="560" height="315" frameborder="0" allowfullscreen="" title="YouTube video player" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin"></IFRAME></P> Tue, 30 Jul 2024 15:45:23 GMT lara_rachidi 2024-07-30T15:45:23Z https://community.databricks.com/t5/databricks-tv/improving-text2sql-performance-with-ease-on-databricks/ba-p/81083 <P><SPAN>How a Databricks team managed to achieve a top score by leveraging few shot learning and fine-tuning the open-source Llama3 8B Instruct model. The Spider benchmark is a widely recognized standard for evaluating the performance of text-to-SQL systems. This post dives into how the authors achieved scores of 79.9% on the Spider development dataset and 78.9% on the test dataset in less than a day of work using the open-source Llama3 8B Instruct model – a remarkable 19-point improvement over the baseline. This performance would place it in a top-10 spot on the now-frozen Spider leaderboard, thanks to strategic prompting and fine-tuning on Databricks. Blog Post: <A href="https://www.databricks.com/blog/improving-text2sql-performance-ease-databricks?utm_source=bambu&amp;utm_medium=social&amp;utm_campaign=advocacy" target="_blank">https://www.databricks.com/blog/improving-text2sql-performance-ease-databricks?utm_source=bambu&amp;utm_medium=social&amp;utm_campaign=advocacy</A></SPAN></P> Tue, 30 Jul 2024 15:45:23 GMT https://community.databricks.com/t5/databricks-tv/improving-text2sql-performance-with-ease-on-databricks/ba-p/81083 lara_rachidi 2024-07-30T15:45:23Z