Feature
- Add support for legacy binary office formats (#3804) (
d0bca90) - msword: Parse native Word charts as classified pictures with data (#3809) (
26ff0b2) - video: Add VideoPipeline, VideoFormatOption, and InputFormat.VIDEO (#3768) (
6f0df37) - asr: Add English-only and Distil-Whisper presets to native backend (#3741) (
39b67b6)
Fix
- Fix the profiling of NemotronOcrModel (#3816) (
ccd09fb) - pptx: Use a safe XML parser (#3831) (
a71c447) - jats: Preserve nested list structure in list items (#3619) (
43f56c4) - Skip empty display formulas in the JATS backend instead of truncating the document (#3813) (
320c517) - Fall back to default image size for non-numeric AsciiDoc dimensions (#3814) (
e398d5a) - vlm: Use each element's own page for force_backend_text (#3807) (
469e1ad) - visualization: Scale top y-edge by scale_y in draw_clusters (#3790) (
0844a97) - Use generation_config from options in PictureDescriptionVlmEngineModel (#3778) (
e548307) - docx: Clean up DrawingML temp dir on conversion error (#3797) (
9b5aabc) - uspto: Detect Grant Full Text Data/XML v2.5 patents case-insensitively (#3801) (
6fb35f9) - pptx: Remove ineffective subtitle label expression (#3785) (
d278116)