LLMs for language and code + Time series and geospatial foundation models
SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion