LLM Wiki

LLM Wiki is a free, open-source tool designed to compile and maintain a structured wiki from raw documents using large language models. It addresses the challenge of synthesizing and organizing research, particularly on transformer architectures and their scaling properties, by automating the creation and upkeep of a knowledge base. The platform is an open-source implementation inspired by Karpathy's LLM Wiki.

The tool operates by ingesting sources such as articles, papers, notes, and transcripts, which serve as the immutable foundation for the wiki. Upon adding a new source, the LLM reads the content, writes a summary, updates relevant entity and concept pages, and flags any contradictions with existing information. This process can affect multiple wiki pages per source, allowing for comprehensive updates and cross-referencing. The generated wiki consists of markdown pages that include summaries, dedicated entity pages, and extensive cross-references, all authored by the LLM. Users interact primarily by reading the wiki and posing complex queries against the synthesized knowledge, rather than re-deriving answers from raw data each time. Notably, valuable answers from these queries are incorporated back into the wiki as new pages, enabling the knowledge base to grow and refine over time.

LLM Wiki also offers linting features that run health checks across the wiki to identify inconsistencies, outdated information, orphaned pages, and missing cross-references. The LLM can suggest new questions to explore and recommend additional sources to ingest, further enhancing the depth and accuracy of the knowledge base. The schema layer, defined by a configuration file, determines the structure of the wiki, the conventions to be followed, and the workflows to be executed during ingestion.

0 license, and its open-source nature allows users to access, modify, and deploy the tool as needed. LLM Wiki is suited for those seeking to automate the organization and synthesis of complex research materials, leveraging LLMs to handle the often tedious aspects of knowledge base maintenance.