Bringing Text-to-SQL to Business Intelligence Production in Large Enterprises

Explore a 38-minute conference talk on bringing Text-to-SQL capabilities to Business Intelligence production in large enterprises. Discover how Tencent bridges the gap between Text-to-SQL tools and natural language querying for both technical and non-technical users. Learn about the process of selecting WizardCoder-34B as the foundational model for fine-tuning, and understand the development of a training instruction set focused on query pattern coverage and business context representation. Gain insights into optimizing inference performance and cost, and examine the evaluation process using GPT4 and the BIRD benchmark. Compare the specialized model's performance against GPT4 in handling complex real-life queries, particularly those requiring table joins. Presented by Hehuan Liu, Algorithm and Data Scientist, and Kun Cheng, Product Manager, both from Tencent.