Teaching PyTorch To Read Your Worst PDFs With Docling
summary
Docling is an MIT‑licensed PyTorch-native library that detects file formats, runs optimized pipelines, and returns a single pydantic canonical document preserving hierarchy, tables, figures and captions across 10+ formats. It runs locally or as a container, offers a CLI and serve API, converts a 200-page PDF in <1s, extracts charts to tables or code, and supports batch scale to billions — a reported run of 918M docs for $35k versus much higher VLM costs.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:24What is Docling
- 3:08Docling Serve modes
- 4:30Conversion vs extraction
- 6:17Docling CLI demo
- 7:40Community growth
- 8:24Nvidia & Hugging Face
- 10:31Cheap scrapers vs VLMs
- 13:54How Docling works
- 15:40Document structure
- 16:44Ecosystem & flexibility
- 17:51Evaluation framework
- 18:34Model innovation
- 20:20Granite Docling
- 21:46Training on ISO standard
- 22:29Additional features and community contributions
- 23:11Closing
related talks
01h 07mMeet Docling: The “Pandas” for Document AI
01h 25mLearn to Unlock Document Intelligence with Open-Source AI
29mConquering PDFs- document understanding beyond plain text
31mDocling: Get Your Documents Ready for Gen AI
01h 19mLearn to Unlock Document Intelligence with Open-Source AI
31mConquering PDFs: document understanding beyond plain text
01h 00mComplex Data Ingestion with Open Source AI
39mAdventures in Understanding Documents