bump(package): update version to 0.6.3 (#110)

fix(structure): fix table cell matching, batch formula inference, and…

… improve markdown output (#100)

* fix(structure): fix table cell matching, batch formula inference, and improve markdown output

- Fix IoA space mismatch in wired table stitching by always using structure cell bboxes; add cross-row OCR deduplication for large cells
- Add predict_images() for cross-page formula batching into a single ONNX inference call, reducing overhead for multi-page documents
- Improve markdown: downgrade ABSTRACT/REFERENCES to h2, require text on both sides for inline formulas, add bullet list formatting, fix paragraph continuation across figures/tables
- Speed up formula preprocessing with bilinear resize (~4x faster)
- Remove premature dedup_by in cluster_positions to match PaddleX
- Use <br/> instead of space for multi-line OCR content in table cells

* fix(structure): per-page error handling, batch chunking, and markdown fixes

feat(vl): add PaddleOCR-VL-1.5 support with text spotting and seal re…

…cognition (#88)

* feat(vl): add PaddleOCR-VL-1.5 support with text spotting and seal recognition

- Add support for PaddleOCR-VL-1.5 model with new tasks: Spotting and Seal
- Update documentation to mention PaddleOCR-VL-1.5 support
- Change huggingface-cli to hf in download commands
- Fix clippy warnings: collapsible if statements, type complexity,
  abs_diff, repeat_n, and needless range loops
- Improve layout detection adapter with PaddleX merge modes

* refactor(layout): replace tuple type alias with named struct and fix row sort order

* fix: address code review feedback for layout and VL modules

* fix: use config score_threshold as fallback for missing class thresholds

feat: implement LightOnOCR model (#81)

* Implement LightOnOCR image processing, text model, and vision model

- Added `processing.rs` for image preprocessing logic, including resizing and normalization.
- Introduced `text.rs` for the LightOnOCR text model, implementing rotary embeddings, attention layers, and MLPs.
- Created `vision.rs` for the Pixtral Vision Model, including patch convolution and transformer layers with rotary embeddings.
- Enhanced utility functions in `utils.rs` to handle character-based substring matching for improved performance with non-ASCII characters.

* feat: Enhance LightOnOCR configuration and processing with error handling and optimizations

* fix: Improve numerical stability in dtype handling for LightOnOCR models

* fix: Improve numerical stability in dtype handling for F16 and BF16

refactor: Simplify path extraction in unclip function

chore: Add workspace configuration to Cargo.toml (#72)

* chore: Add workspace configuration to Cargo.toml

* refactor: Simplify conditional parsing for builder config attribute

* refactor: Update Cargo.toml to use workspace attributes for package metadata

Add unirec-0.1b model (#60)

* Add utility functions for VL models and document parsing

- Implement error conversion functions for Candle to OCRError.
- Add tensor manipulation functions including rotation and concatenation.
- Create Markdown conversion functions for layout elements.
- Implement OTSL to HTML table conversion with support for various tags.
- Add image processing helpers for cropping margins from images.
- Introduce functions for calculating bounding box areas and overlap ratios.
- Implement detection and filtering of overlapping boxes in layout detection results.
- Add tests for utility functions to ensure correctness.

* refactor: Update usage documentation and examples for improved clarity and consistency

* fix: Remove unnecessary reference for ignore_labels in markdown conversion

* refactor: Optimize sinusoidal embedding generation for GPU efficiency

Add PaddleOCR-VL document parsing and image processing modules (#52)

* Add PaddleOCR-VL document parsing and image processing modules

- Implemented `PaddleOcrVlDocParser` for layout-first document parsing using PP-DocLayoutV2 and PaddleOCR-VL.
- Introduced configuration struct `PaddleOcrVlDocParserConfig` to customize parsing behavior.
- Added functions for bounding box manipulation, layout element sorting, and order assignment.
- Created `PaddleOcrVlImageInputs` and `preprocess_images` function for image preprocessing, including smart resizing and normalization.
- Implemented table output post-processing with HTML conversion and token parsing.
- Added unit tests for smart resizing and image preprocessing outputs.

* bump version to 0.3.2

* Enhance layout detection and processing for PaddleOCR-VL

- Update output handling to support reading order for PP-DocLayoutV2.
- Introduce `is_reading_order_sorted` flag in layout detection output.
- Refactor document parsing to utilize reading order when available.
- Add new configuration parameters for image processing.
- Improve position embedding interpolation in vision model.

* Refactor layout post-processing types and update Clippy linter configuration

* Refactor error handling and improve layout post-processing in PaddleOCR-VL

* Bump version to 0.4.0 and update HTML entity comment in image processing

chore: Update version to 0.3.1 and refine tokenizers dependency confi…

…guration (#51)

Add documentation for pre-trained models and usage guide (#48)

* Add documentation for pre-trained models and usage guide

- Created models.md to document available pre-trained models for OCR and document understanding, including details on text detection, recognition, character dictionaries, preprocessing, and document structure models.
- Added usage.md to provide a comprehensive guide on using OAROCR, covering basic OCR pipeline, batch processing, builder APIs for text recognition and document structure analysis, GPU acceleration, and configuration options.

* refactor: Update model file paths in usage documentation for consistency