WeVisDoc-4B

Use it to parse document page images into Markdown on a local machine or through a self-hosted vLLM service.

Availability

Checkpoint linked on Hugging Face; local Transformers and vLLM instructions provided

Pricing model

Not established

Deployment

Local inference or vLLM service

Privacy

Not established

Licence

Apache-2.0

• WeVisDoc-4B turns a page image into structured Markdown, with LaTeX formulas and HTML tables. ^[2]
• The WeVisDoc-4B checkpoint is linked on Hugging Face.
• The instructions support local Transformers inference and serving with vLLM.
• WeVisDoc-4B scores 95.38 Overall on OmniDocBench v1.6. ^{[1] [2]}

  WeVisDoc-4B achieves an Overall score of 95.38 on OmniDocBench v1.6 and a mean Overall score of 75.54 across the three PureDocBench tracks, ranking first among the compared end-to-end parsers in all four settings.

• WeVisDoc-4B has a 75.54 mean Overall score across the three PureDocBench tracks. ^{[1] [2]}

  WeVisDoc-4B achieves an Overall score of 95.38 on OmniDocBench v1.6 and a mean Overall score of 75.54 across the three PureDocBench tracks, ranking first among the compared end-to-end parsers in all four settings.