RAGFlow v0.10 introduces the Text2SQL feature in response to community demand. Traditional Text2SQL requires model fine-tuning, which can significantl

Implementing Text2SQL with RAGFlow v0.10

submited by

Style Pass

2024-09-23 08:00:09

RAGFlow v0.10 introduces the Text2SQL feature in response to community demand. Traditional Text2SQL requires model fine-tuning, which can significantly increase deployment and maintenance costs when used in enterprise settings alongside RAG or Agent components. RAGFlow’s RAG-based Text2SQL leverages the existing (connected) large language model (LLM), enabling seamless integration with other RAG/Agent components without the need for additional fine-tuned models.

General speaking, you need to prepare a knowledge base for generating Text2SQL prompts, which contains various examples of natural language being converted to SQL statements. A user query is first sent to this knowledge base to retrieve similar examples. The retrieved examples are then concatinated into prompts for the LLM to generate the final SQL statement. The generated SQL is used directly to query the database. If the returned result is incorrect or if, even worse, nothing is retrieved, the generated SQL will be considered incorrect, and the LLM will be called again to regenerate a SQL statement until the predefined upper limit is reached.

Therefore, Text2SQL relies on multiple rounds of orchestration. RAGFlow v0.10.0 encapsulates this Text2SQL feature into a convenient, built-in Agent component. In upcoming releases, we plan to adjust this workflow. The goal is to enable users to manually add or update text2SQL examples in the knowledge base, as indicated by the dashed arrow above.