Lmql

LMQL Overview

LMQL, or Language Model Query Language, is an innovative programming language designed specifically for interacting with language models. It allows users to supercharge their prompting with a range of features including constraints, a debugger, decoders, transformers, templates, retrieval, interaction, distributions, token masking, and control flow. LMQL is a project by the SRI Lab at ETH Zürich and is available in a web-based Playground IDE or can be installed via the Python package manager. It supports high-level constraints, regular Python control-flow, and even async functions, enabling users to query external web services during decoding. LMQL also supports OpenAI Chat models, including ChatGPT and GPT-4.

LMQL Highlights

• LMQL allows users to express natural language prompts that also contain code, enabling them to steer the language model's reasoning process.
• Users can specify high-level, logical constraints over the language model output. These constraints are automatically converted into token-level prediction masks, which can be enforced eagerly during text generation.
• LMQL includes a Playground IDE for query development, enabling users to inspect the interpreter state, validation result, and model results at any point during generation.