PaDoc proposes parallel decoding for document parsing
The arXiv paper describes a layout-grounded MLLM parser that decodes document regions concurrently.
Why it matters
The work targets a core bottleneck in end-to-end document parsing: long autoregressive decoding paths for independent page regions. If validated beyond the reported benchmark, the approach could make MLLM-based document parsing faster while preserving page-level context.
The key points
- 1.PaDoc decodes layout and regional content branches concurrently.
- 2.The method uses shared full-page representation inside a single MLLM.
- 3.Reported OmniDocBench Full Overall layout F1 is 91.1.
Researchers proposed PaDoc, a layout-grounded document parser that treats a predicted page layout as a branching structure over a shared page representation. The method is designed to keep full-page context while decoding layout and regional content branches concurrently, using packed variable-length ancestor attention and masked parallel decoding in a single MLLM. On OmniDocBench Full, the reported Overall layout F1 is 91.1.
⚡ Try this today
Read the paper before building high-throughput document parsers that currently serialize page regions autoregressively.
Sources & original reporting
This brief summarizes and links to reporting from the publishers below.
Enjoyed this brief? Get the next one in your inbox.
More in Research
New studies probe spatial reasoning in vision models
SpaRRTa, SMA and PinpointQA target gaps in model spatial understanding for embodied AI.
CW-BASS v2 targets pseudo-label filtering with DINOv2 teachers
The arXiv paper proposes a saturation-aware method for semi-supervised semantic segmentation.
Study tracks ChatGPT Enterprise use across organizations
The paper links account records to roles, tasks and public-company data through March 2026.