Large Language Models have taken the Artificial Intelligence community by storm. Their recent impact has helped contribute to a wide range of industries like healthcare, finance, education, entertainment, etc. The well-known large language models such as GPT, DALLE, and BERT perform extraordinary tasks and ease lives. While DALLE 2 can create images responding to a simple textual description, GPT-3 can write an excellent essay, complete codes, summarize long textual paragraphs, answer questions like humans, and generate content given just a short natural language prompt. These models are helping Artificial Intelligence and Machine Learning move rapidly through a paradigm shift.
Recently, a team of researchers has introduced LMQL, an open-source programming language, and platform for language model interaction. LMQL, which stands for Language Model Query Language, improvises the capabilities of Large Language Models (LLMs) by combining prompts, constraints, and scripting. Being a declarative, SQL-like language based on Python, LMQL extends static text prompting with control flow, constraint-guided decoding, and tool augmentation. With this type of scripting, LMQL simplifies multi-part prompting flows with a very small piece of code.
The researchers have used LMQL to enable LMP (Language Model Programming), which generalizes language model prompting from pure text prompts to a combination of text prompting and scripting. LMQL influences the constraints and control flow from an LMP prompt to generate an efficient inference procedure. These super logical and high-level constraints are translated to token masks with the help of some evaluation semantics that is keenly enforced at the time of generation.
The team has introduced LMQL to avoid the high cost of re-querying and validating generated text. This can help LMQL produce text closer to the desired output on the first attempt without needing subsequent iterations. Also, LMQL constraints allow users to guide or steer the text generation process according to their desired specifications, like ensuring that the generated text follows certain grammatical or syntactic rules or that certain words or phrases are being avoided.
The researchers have mentioned how LMQL can capture a wide range of state-of-the-art prompting methods, such as interactive flows, that are difficult to implement with existing APIs. The evaluation shows that LMQL retains or improves the accuracy on numerous downstream tasks while significantly reducing computation or cost in pay-to-use APIs, resulting in 13-85% cost savings.
LMQL allows users to express a wide range of common and advanced prompting techniques simply and concisely. It integrates with the Hugging Face’s Transformers, OpenAI API, and Langchain. The developer resources for the same are available at lmql.ai, and a browser-based Playground IDE is available for experimentation.
To summarize, LMQL seems like a promising development as the evaluation demonstrates how LMQL is a powerful tool that can improve the efficiency and accuracy of language model programming. It can make it easier for users to achieve their desired results with fewer resources.
Check out the Tool. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 18k+ ML SubReddit, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.
Tanya Malhotra is a final year undergrad from the University of Petroleum Energy Studies, Dehradun, pursuing BTech in Computer Science Engineering with a specialization in Artificial Intelligence and Machine Learning.
She is a Data Science enthusiast with good analytical and critical thinking, along with an ardent interest in acquiring new skills, leading groups, and managing work in an organized manner.