Lexoid

One commit at a time, we are keeping multimodal document understanding, parsing, and information extraction free under a permissive license. If you like the idea, or if something is missing or something we can do better, give us a nudge. https://github.com/oidlabs-com/Lexoid Initial benchmarks: https://oidlabs-com.github.io/Lexoid/benchmark.html#benchmark-results Work in progress ...

Nice! This should be helpful. https://github.com/oidlabs-com/Lexoid?tab=readme-ov-file#ben...

Why we built Lexoid: 1. Recent advancements in LLMs make them incredibly powerful for extracting meaning from unstructured documents. We wanted to utilize these different LLMs through a single library. 2. Static (a.k.a. non-LLM-based) parsing libraries like PyMuPDF work fast and accurately for simple documents, but often their licenses aren’t as permissive as we’d like. 3. We wanted a single library that could integrate both static and LLM-based approaches in a seamless and efficient way. Key Features: - Unified Parsing: Use either static parsers or LLM-based ones, depending on your needs. - Auto-routing Mode: Pages are analyzed independently—faster static parsers handle straightforward pages, while LLMs process pages requiring deeper understanding. Why we think it’s useful: Static parsers are fast but struggle with complex, unstructured