v2.130.0
tanchongmin/strictjsonv2.130.0Sep 22, 2026by docling-ops[bot]
Upgrading?
AI Summary
Adds Mineru 2.5 Pro VLM support and fixes chart extraction double registration. Includes numerous improvements for Markdown table parsing, PDF, DOCX, and HTML backends.
Key Highlights
- Add Mineru 2.5 Pro to the supported VLMs
- Honor backslash-escaped pipes in Markdown tables
- Fix double preset registration of chart extraction
- Omit only unset arguments in service client
- Harden the LibreOffice profile and timeout kill
New Features
- Mineru 2.5 Pro VLM support
- Recognize .markdown files
- Defer docling-parse import in PDF backend
- Emit <caption> of a table in HTML
- Fix VLM end token stripping
Full Release Notes
### Feature * Add Mineru 2.5 Pro to the supported VLMs ([#4322](https://github.com/docling-project/docling/issues/4322)) ([`9a2c3e7`](https://github.com/docling-project/docling/commit/9a2c3e775051f2e19de7f70b04b7241ac95c70fc)) ### Fix * **service client:** Omit only unset arguments but keep values set equal to defaults ([#4349](https://github.com/docling-project/docling/issues/4349)) ([`35c6af5`](https://github.com/docling-project/docling/commit/35c6af53b58849aafbf491968fd68363cdf732ff)) * Remove double preset registration of chart extraction ([#4348](https://github.com/docling-project/docling/issues/4348)) ([`878831c`](https://github.com/docling-project/docling/commit/878831c786cf2ac5b0d726a8cb30906b84d09a2d)) * **md:** Honor backslash-escaped pipes in Markdown tables ([#4313](https://github.com/docling-project/docling/issues/4313)) ([`52e52fd`](https://github.com/docling-project/docling/commit/52e52fd738fc04f73b4566d2a4ce0dee7cbd848e)) * Fix the VLM end token stripping. Extend the unit tests ([`7748f6e`](https://github.com/docling-project/docling/commit/7748f6e581c366ac39398f50a12b805ae8e018d6)) * **docx:** Keep list items when a list starts above indent level 0 ([#4188](https://github.com/docling-project/docling/issues/4188)) ([`b078ea3`](https://github.com/docling-project/docling/commit/b078ea34921eafe181141e6179e87885196497c3)) * **backend:** Defer the docling-parse import in the PDF backend ([#4286](https://github.com/docling-project/docling/issues/4286)) ([`1b40928`](https://github.com/docling-project/docling/commit/1b4092810c364e1fdc0b5d358faf353f9ca9447e)) * **html:** Emit the <caption> of a table and link it to the TableItem ([#4289](https://github.com/docling-project/docling/issues/4289)) ([`d6d9da0`](https://github.com/docling-project/docling/commit/d6d9da076a4f87da969af109bfa89e5d2a097a1e)) * Force eager attention for Nemotron Parse v2 (#4307) ([#4328](https://github.com/docling-project/docling/issues/4328)) ([`54c7cf2`](https://github.com/docling-project/docling/commit/54c7cf2e3f7217c16525eb9fb7001a9fb890af5c)) * **ebcdic:** Keep fixed-point notation for scaled numbers ([#4296](https://github.com/docling-project/docling/issues/4296)) ([`244042a`](https://github.com/docling-project/docling/commit/244042af9d97bfb73e31883698857def94056db8)) * **docx:** Don't reuse a list group across an intervening table ([#4305](https://github.com/docling-project/docling/issues/4305)) ([`4141a99`](https://github.com/docling-project/docling/commit/4141a9963a2537f83146be6ed240c46ad16cafa5)) * **afp:** Follow PTOCA control sequence chaining ([#4297](https://github.com/docling-project/docling/issues/4297)) ([`df6b341`](https://github.com/docling-project/docling/commit/df6b341c8b4de1e88e2d6188830cf36ae0534e6d)) * **pptx:** Record slide positions as TOPLEFT, the origin python-pptx reports ([#4294](https://github.com/docling-project/docling/issues/4294)) ([`dd05464`](https://github.com/docling-project/docling/commit/dd05464edf7ba736f732be6aaa6dbf996419b111)) * **office:** Harden the LibreOffice profile, flags, and timeout kill ([#4228](https://github.com/docling-project/docling/issues/4228)) ([`14789f2`](https://github.com/docling-project/docling/commit/14789f2c2c542f8d127a0aac85eb779d5bba554e)) * **docx:** Clamp heading levels to the OOXML 1-9 range ([#4319](https://github.com/docling-project/docling/issues/4319)) ([`1c2dbec`](https://github.com/docling-project/docling/commit/1c2dbec06fb6d8f55885bd2b135bbe0484fb175c)) * **xlsx:** Avoid duplicate sparse table fragments ([#4302](https://github.com/docling-project/docling/issues/4302)) ([`891de07`](https://github.com/docling-project/docling/commit/891de07317f4bff5769ed7a46882ecc4d75797e8)) * **email:** Fall back to the HTML body when the plain part is blank ([#4295](https://github.com/docling-project/docling/issues/4295)) ([`8e37971`](https://github.com/docling-project/docling/commit/8e37971218f2af9184273b692cf255466f149bca)) * Skip already downloaded EasyOCR models unless force is set ([#4317](https://github.com/docling-project/docling/issues/4317)) ([`8522eb7`](https://github.com/docling-project/docling/commit/8522eb7553acf896240f4a0e39c3ee9f4d85125f)) * **backend:** Defer pypdfium2 import in office and LaTeX backends ([#4285](https://github.com/docling-project/docling/issues/4285)) ([`abca4be`](https://github.com/docling-project/docling/commit/abca4be4d1de5fc6b89062250f6b31cb6d1fcb22)) * Recognize .markdown files ([#4306](https://github.com/docling-project/docling/issues/4306)) ([`557f0fa`](https://github.com/docling-project/docling/commit/557f0fa60f67a30030fd5c8392fb47470a6d6e81)) ### Documentation * **ocr:** Fix the dead same-page anchor in OCR.md ([#4310](https://github.com/docling-project/docling/issues/4310)) ([`6e22317`](https://github.com/docling-project/docling/commit/6e22317ca3217ef26a95648160dff16146767ba5)) * **examples:** Fix stale paths in the VLM pipeline example comments ([#4311](https://github.com/docling-project/docling/issues/4311)) ([`490fa64`](https://github.com/docling-project/docling/commit/490fa646e2b73af1ad5b506c434549c8b75dd9af))
Related Repositories
rasbt/LLMs-from-scratch★ 107,043
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
lyogavin/airllm★ 40,153
AirLLM 2.8T LLM inference with single 4GB GPU
NirDiamant/RAG_Techniques★ 29,958
This repository showcases various advanced techniques for Retrieval-Augmented Generation (RAG) systems. Each technique has a detailed notebook tutorial.
microsoft/OmniParser★ 25,539
A simple screen parsing tool towards pure vision based GUI agent
karpathy/nn-zero-to-hero★ 24,906
Neural Networks: Zero to Hero