HPD-Parsing reaches 4,752 tokens/s by replacing page-wide autoregression
PaddlePaddle · hf · 2026-07-22
HPD-Parsing replaces page-wide autoregression with hierarchical parallel decoding
This paper tackles the sequential bottleneck in unified VLM-based document parsers.
- Existing parsers process the whole page jointly, but still generate outputs through one token-by-token autoregressive path, which becomes a bottleneck as documents get longer.
- The authors argue that document parsing has two different needs: global layout analysis and parallel block-level content parsing.
- Their method, HPD-Parsing, uses a main layout branch to organize the page structure and dynamically assigns block content to concurrent decoding branches.
- They further add progressive multi-token prediction (P-MTP) to cut the number of decoding steps inside each branch.
- On public benchmarks, the system reaches 4,752 tokens/sec, which is 2.62× faster than the fastest prior document parsing model and 3.06× faster than the vanilla autoregressive baseline.
- Accuracy remains competitive, suggesting hierarchical parallel decoding is a viable alternative to full-page autoregression for document parsing.
More from Research
- RoboMME benchmarks long-horizon memory across 16 robot tasks — micoolcho · 2026-07-23
- NeuroAI review finds vision far ahead of language after a decade of brain–DNN work — mtoneva1 · 2026-07-23
- Mila Quebec will host a talk on what AI benchmarks really measure for African languages — hugo_larochelle · 2026-07-23
- An agent stack diagram says production AI is 90% architecture, not prompts — theomitsa · 2026-07-23
- MIT’s free “SLAM for Dummies” guide turns robotics navigation into a hands-on tutorial — lukas_m_ziegler · 2026-07-23
- Sol 5.6 reportedly writes a full research paper from one prompt — conitzer · 2026-07-23