Starstruk Logo MarkStarstruk

v2.128.0

0xSojalSec/airllmv2.128.0Sep 16, 2026by docling-ops[bot]

Upgrading?

What changed from v2.10.1 → v2.128.0?What's new in v2.129.0 after v2.128.0?

AI Summary

This release adds minimal support for AFP documents and Nvidia Nemotron Parse v2 as a VLM model. It also refactors the docling-parse backend to remove pypdfium.

Key Highlights

  • Added minimal AFP document support
  • Added Nvidia Nemotron Parse v2 as a VLM model
  • Refactored PDF backend to remove pypdfium
  • Improved image EXIF orientation handling
  • Enhanced Chandra OCR parsing

New Features

  • AFP document support
  • Nvidia Nemotron Parse v2 VLM model

Full Release Notes

### Feature

* Add minimal AFP document support ([#4177](https://github.com/docling-project/docling/issues/4177)) ([`1cdae27`](https://github.com/docling-project/docling/commit/1cdae2749e89853a0c6d7896ced6cdfddcb76721))
* Add nvidia/NVIDIA-Nemotron-Parse-2.0 as vlm model ([#4253](https://github.com/docling-project/docling/issues/4253)) ([`01b5738`](https://github.com/docling-project/docling/commit/01b57389a33c5b2eac974b41344fb957a1e80838))
* **pdf:** Refactor the docling-parse backend to remove pypdfium ([#4244](https://github.com/docling-project/docling/issues/4244)) ([`ca0ecd4`](https://github.com/docling-project/docling/commit/ca0ecd413e5902bb423d5224b02969edf2561971))

### Fix

* **asciidoc:** Block titles, picture parenting, skipped headings, and list dedent ([#4171](https://github.com/docling-project/docling/issues/4171)) ([`7fe1b3c`](https://github.com/docling-project/docling/commit/7fe1b3c602105b2f3fb7b6a75d0c529e4a4da4b2))
* **image:** Apply the EXIF orientation when loading image frames ([#4247](https://github.com/docling-project/docling/issues/4247)) ([`77ab16d`](https://github.com/docling-project/docling/commit/77ab16d8a6510572d8c720d3df2ad1bc99c7fd92))
* Improve Chandra OCR parsing and preserve document structure ([#4239](https://github.com/docling-project/docling/issues/4239)) ([`b322564`](https://github.com/docling-project/docling/commit/b32256402addc02452697805ee25f38b5e8abffb))
* **email:** Keep rendered display names parseable and single-line ([#4242](https://github.com/docling-project/docling/issues/4242)) ([`e8c6092`](https://github.com/docling-project/docling/commit/e8c6092fe6ee995ce0719a71352b9f1fad5350cf))
* **docx:** Sanitize fragment-only relationship targets in DOCX archives before parsing ([#4243](https://github.com/docling-project/docling/issues/4243)) ([`1aec03f`](https://github.com/docling-project/docling/commit/1aec03f1c3326584e8a7b97b2d0ed2ad95d946b4))

### Documentation

* Fix stale TransformersModelType enum member in vision models guide ([#4251](https://github.com/docling-project/docling/issues/4251)) ([`bbdaeda`](https://github.com/docling-project/docling/commit/bbdaeda1d8a749b9dafe16d701cbacf495435c11))

Related Repositories

rasbt/LLMs-from-scratch★ 107,048

Implement a ChatGPT-like LLM in PyTorch from scratch, step by step

lyogavin/airllm★ 40,162

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