Release v1.2.0: Dense Markdown & Zero Line-Spacing
This release introduces aggressive text post-processing to make our output perfectly align with the dense output style of Microsoft's markitdown.
New Features & Improvements
- Aggressive Empty Line Elimination: We've implemented a rigorous filtering pass on the final generated Markdown to eliminate all blank, empty, or whitespace-only lines.
- 1:1 Microsoft MarkItDown Parity: The line spacing in our extracted documents now matches the zero-spacing density of Microsoft's native implementation, significantly reducing unnecessary file bloat and line counts.
- Global Support: This text-density filtering applies seamlessly across the board, whether the text is extracted natively from text/PDFs or generated through the Moondream2 GPU Vision/OCR pipeline.
- Zero Overhead: The text-filtering logic is handled entirely on the host side, ensuring 0 performance overhead or latency during the core CUDA/VLM inference.
🔧 Fixes & Refactoring
- Refactored
sanitize_outputinside the parsing layer. - Bumped dependencies and version configurations for v1.2.0.