LMQL

LMQL is an innovative programming language which combines language model prompting with scripting to enable the successful completion of a wide range of tasks. LMQL allows users to specify constraints over the language model output, making it easier to create a model which is adapted for individual tasks. The language utilizes efficient inference procedures to minimize the expensive calls to the language model. LMQL can capture many state-of-the-art prompting methods to create interactive flows that are challenging to implement with existing high-level APIs. An evaluation of LMQL has shown that it is successful in retaining or increasing the accuracy of a number of downstream tasks at the same time as significantly reducing the required amount of computation. These results have been validated with a 26-85% cost savings in the case of pay-to-use APIs. In summary, LMQL is a beneficial programming language for large language models which provides users with a simple interface to incorporate scripting functionality and enables constraints to be specified over the language model output, resulting in an efficient inference procedure which minimizes the number of expensive calls to the language model. Furthermore, LMQL can capture a wide range of state-of-the-art prompting methods, ultimately reducing the required amount of computation