Extreme LLM, abbreviated as xLLM, relies on multiple specialized large language models, one per top category, to deliver highly relevant answers to specific questions, covering the entire human knowledge or targeted content such as corporate repositories. The user, in addition to the classic prompt, is invited to select or guess top categories.
Behind the scenes, it involves one simple LLM per top category and a reconstructed granular taxonomy of the input sources (crawled webpages, or parsed data). Each LLM has its own set of summary tables: embeddings, links (URLs), dictionary, stopwords, synonyms, n-grams, related content, and so on, to further enrich the answers to user queries. Expert answers are formatted as itemized lists with relevancy score attached to each item, rather than wordy English sentences. It is designed for busy professionals, researchers or scientists who know what they are looking for.
The default parameters are selected based on usage rather than training: users are allowed to choose parameters that best fit their needs; default values are the most popular choices. It leads to self-tuning, but also to customized output. Without neural networks or training, the app is very fast and requires much less data. Yet it delivers better results by privileging the quality of input sources over quantity, and by extracting only the essential material including detected structures.
Well documented and available as open source (thus, free), it is also very frugal in terms of GPU, cloud or bandwidth usage. In short, cheaper than vendor solutions by several orders of magnitudes.
Current Version
As any open-source project, it is a work in progress. It has been tested on the largest and best math repository (Wolfram), with developers currently adding Wikipedia and content from books parsed in their native format (LaTeX). I am also working with a fortune 100 company to implement a version for corporate needs, and with a number of startups.
The approach is radically different from OpenAI and the likes, with too many foundational features and enhancements to list in this short introduction. It is entirely home-made from scratch, using well-thought-out methods and algorithms, avoiding many issues found in standard libraries such as NLTK.
Over the last few months, I posted datasets and code on GitHub, and wrote several papers on the topic. The purpose of this article is to share detailed and unified documentation, to allow AI companies and clients to easily implement the technology. This is just the very starting point of a long journey, to make GenAI accessible to a large audience, at essentially no cost, and deliver actual and measurable value.
Sources and Documentation
The detailed description is included in my project textbook, available on GitHub, here. The Python code and datasets are in the same GitHub repository. Future updates will be added to the same document and repository, at the same location. I am currently working on a free Web API. Funding is not an issue.
Note that the project textbook (still under development) contains a lot more than xLLM. The reason to share the whole book rather than just the relevant chapters is because of cross-references with other projects. Also, clickable links and other navigation features in the PDF version work well only in the full document, on Chrome and other viewers, after download. The very core is section 7.2.2. I highly recommend that you start with this section as it links to everything else if you want a deep dive. Excluding the Python code, that part is only 3 pages long.
To not miss future updates on this topic and GenAI in general, sign-up to my newsletter, here. Upon signing-up, you will get a code to access member-only content. There is no cost. The same code gives you a 20% discount on all my eBooks in my eStore, here.
Author
Vincent Granville is a pioneering GenAI scientist and machine learning expert, co-founder of Data Science Central (acquired by a publicly traded company in 2020), Chief AI Scientist at MLTechniques.com and GenAItechLab.com, former VC-funded executive, author (Elsevier) and patent owner — one related to LLM. Vincent’s past corporate experience includes Visa, Wells Fargo, eBay, NBC, Microsoft, and CNET. Follow Vincent on LinkedIn.