Feature
- pdf: Run the heading style signal without generate_parsed_pages and add --heading-hierarchy (#5071) (
6a12963) - ocr: Replace broken PDF text layers with OCR (#4654) (
b3b2cd9)
Fix
- vlm: Drop the spaces around Chandra br line breaks (#4845) (
7814a4e) - asr: Import torch lazily in the transformers ASR decoder (#5072) (
46504cc) - asr: Forward temperature and anti-hallucination options to mlx-whisper (#5067) (
8bb0b6c) - Recognise Symbol and Wingdings private-use bullets as list markers (#5043) (
26ef35a) - Skip macOS metadata members in ZIP inputs (#5026) (
da9e947) - vlm: Ignore whitespace between Chandra list paragraphs (#5049) (
2fa79ef) - image: Treat zero DPI metadata as missing (#4915) (
2a55ea6) - pdf: Normalize Arabic presentation forms in extracted text (#3929) (
00352a1) - reading-order: Read heads split by a spanning block top-down (#4438) (
ecdba67) - ocr: Stop concurrent conversions from crashing the tesserocr engine (#4941) (
868768f)