NeoMME: an efficient Multimodal-native and Multilingual Encoder
Source
huggingface.co
Date
Context
NeoMME-Retriever searches document pages as images, preserving tables, charts and layout that text extraction can lose. Its encoder processes image patches and text through one shared network, then produces two search representations in a single pass: a compact vector for finding likely pages and detailed vectors for matching query words to particular page regions. This lets a large document collection use fast initial retrieval followed by finer ranking of the candidates. The release article describes that workflow without requiring a text-generating model to perform the search.
Terms in this piece · Glossary
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
