Docling Files

Feature

• Add nvidia/Nemotron-3.5-Parse as supported VLM model (#4897) (2d1d814)
• cli: Generic engine options for docling convert (#4729) (f57160e)
• service: Allow artifact downloads from trusted private origins (#4892) (8df5b6e)
• forms: Map form fields with a geometric solver when AcroForm widgets are available (#4496) (2f13b3d)
• asciidoc: Keep comments in the notes content layer (#4663) (ba458a2)

Fix

• jats: Convert the content of sub-articles (#4826) (d01c9c0)
• Flatten transparent pictures onto white before enrichment models (#4776) (0c87ce6)
• models: Warn on unpinned revisions for trust_remote_code model specs (#4227) (ae84464)
• docx: Read styles with the invalid type "list" (#4831) (0fe65b2)
• md: Keep a heading or list item without text from taking a later block's text (#4870) (145d379)
• vlm: Avoid PIL warning on palette P+tRNS images (#4392) (2bdc46d)
• iwork: Apply page_range to Keynote slides (#4871) (e84e9aa)
• docx,iwork: Put a sub-list under the list item above it (#4863) (d54d4b8)
• msexcel: Read the series of every chart plot (#4812) (399b983)
• latex: Convert accent and letter macros in body text (#4517) (b845ce6)
• service: Connect artifact downloads to the validated address (#4889) (97dfad6)
• odf: Keep the speaker notes of ODP slides in the notes layer (#4869) (0c75567)

Documentation

• Fix dead LlamaIndex DoclingReader guide link (#4885) (7a25855)

Performance

• odt: Index automatic styles for style lookups (1.6x-3.2x on Word-exported ODT) (#4881) (0b62ddd)
• docx: Cache style chains and reuse paragraph content (1.6x faster) (#4867) (7bb935a)
• docx: Memoize python-docx style lookups (#4862) (25a644d)