From the model card
Fields below are copied from the tags and counters on the HuggingFace repository hantian/layoutreader at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- hantian
- Pipeline tag
- token-classification
- Library
- Transformers
- Framework tags
- PyTorch
- Weight formats
- safetensors
- License tag
cc-by-nc-sa-4.0— read the license file in the repo before relying on it- Downloads (HF counter at last fetch)
- 374,884
- Likes (HF counter at last fetch)
- 45
- Model card
- https://huggingface.co/hantian/layoutreader
Use cases
- Predicting reading order of tokens in scanned PDF or image documents
- Pre-processing step for document OCR pipelines before downstream NLP
- Structuring table or multi-column layouts into sequential text streams
- Evaluating layout-aware token ordering in document understanding benchmarks
Pros
- Targets a specific and well-defined task (reading order prediction) rather than a general-purpose model
- Built on LayoutLMv3, a well-studied document understanding architecture with strong layout encoding
- PyTorch and safetensors weights are provided, covering common training and inference frameworks
- Over 530K downloads reflects substantial real-world adoption in document processing workflows
Cons
- CC-BY-NC-SA-4.0 license prohibits commercial use — a significant constraint for production document pipelines
- Reading order prediction quality degrades on non-standard layouts such as infographics or forms with complex nesting
- No arxiv paper linked in model tags, limiting ability to independently assess methodology
- LayoutLMv3 requires both text tokens and bounding box coordinates as input — not compatible with plain-text pipelines
- Model is task-specific and cannot be repurposed for general document Q&A without additional fine-tuning
Tags
transformerspytorchsafetensorslayoutlmv3token-classificationlicense:cc-by-nc-sa-4.0endpoints_compatibleregion:us