Skip to main content

Overview

By getting language models to output their reasoning as a structured table, we can improve their reasoning capabilities and the quality of their outputs. This is known as Tabular Chain Of Thought (Tab-CoT). We can implement this using Instructor with a response model ensuring we get exactly the data that we want. Each row in the table is represented as a ReasoningStep object.

Example

Sample Output

References

  1. Tab-CoT: Zero-shot Tabular Chain of Thought (https://arxiv.org/pdf/2305.17812)