Feature
Fix
- cli: Exit with code 1 when a document fails to convert (#4888) (
787ad5e) - nemotron-parse: Bound work on untrusted model output (#4923) (
3b0f25f) - Align transformer engine input dtype with model weights (fixes #3110) (#4158) (
a401629) - asr: End merged sentences at full-width CJK punctuation (#4836) (
d18c25b) - Adapt to docling-core 2.103.1 image and HTML changes (
4e6a615) - deps: Require docling-core 2.103.1 (
197a65c) - vlm: Keep every markup heading depth distinct across VLM mappers (
73b7c97) - vlm: Share the Nemotron 3.5 formula delimiter normalizer (
d2b076a) - Normalize Nemotron formula delimiters before export (
91332c9) - Preserve VLM section heading depth in exports (
fec45d3) - Preserve multiline DeepSeek and Unlimited OCR blocks (
62eeaee) - ebcdic: Report a record length shorter than its layout (#4961) (
6470556) - odt: Extract page headers and footers (#4960) (
e3d4743) - Handle truncated PNG chunks in DOCX pictures (#4946) (
10054ec) - latex: Backend updates for better parsing of arxiv papers (#4817) (
29caea8) - otsl: Keep cells when an unclosed cell precedes a closed one (#4916) (
7b9a6b1) - Support Pillow 13 and integer PNG images (#4929) (
ed66a3c) - datamodel: Read one content-sniff window for a path and a stream (#4924) (
ffa6a84)
Documentation
- vlm: Correct the heading rule docs and comments (
91a6439)