← Back to live feed1 story
The new multilingual encoders allow users to search document pages including charts and tables without performing an optical character recognition step. These models integrate with the Sentence Transformers library through a MultiVectorEncoder to process text and images simultaneously. H Company released the tool in two sizes with 260M and 800M parameters.
Day-0 compatibility with the Sentence Transformers framework facilitates immediate fine-tuning and deployment for multimodal applications. This integration provides a standardized way for developers to implement high-efficiency retrieval for a variety of text-and-image datasets in the wild.
Sign in to suggest edits