Surya Order is a vision-encoder-decoder model hosted on Hugging Face that predicts the reading order of elements in document images. It forms part of the Surya OCR project and addresses the need to reconstruct logical text flow from layouts such as multi-column pages, forms, or tables.
The model is provided in Safetensors format with 0.1 billion parameters and uses I64 and F32 tensor types. It carries a cc-by-nc-sa-4.0 license. Users load it through the Transformers library by importing AutoTokenizer and AutoModel, then calling from_pretrained on the repository vikp/surya_order with an optional device_map parameter.
Downloads reached 209070 last month. The model card lists two Spaces that incorporate it and notes that no inference provider has deployed it yet. It is delivered exclusively as a model repository on the Hugging Face platform and can be run in local environments, Colab notebooks, or Kaggle.
The repository page supplies example code for direct loading and links to relevant notebooks. No additional capabilities, integrations, or deployment options are stated.
In the Other AI space, Surya Order takes a focused approach. It focuses on determining the correct reading order of text and elements within complex document layouts for accurate OCR processing. It is built as an open-source project for developers. Surya Order is open source under the Apache-2.0 license. It runs on the web and API.
vikp builds and maintains Surya Order, and the product first shipped in 2024. Development happens publicly on GitHub with 21.1k stars and 51 commits in the last 90 days.
Latest indexed changes and source events
vikp/surya_order verified by the PulseGate indexer
Other apps tracked under the same category.
