GitLab

- Replace complex device selection logic with a single line using torch.device
- Remove redundant checks and imports for better readability and maintainability

- Remove supports_bfloat16 variable and related checks
- Remove model.bfloat16() call for LayoutLMv3ForTokenClassification
- Simplify device selection logic

- Implement full_to_half function to convert full-width characters to half-width
- Apply conversion to span content before merging paragraphs
- Improve text processing for better readability and consistency

- Move batch model initialization outside the loop
- Collect page dimensions before analyzing- Update page info dictionary structure
- Add null dimensions for non-analyzed pages

- Introduce span_block_type_compatible function to check compatibility between span and block types
- Update fill_spans_in_blocks function to use the new compatibility check
- Improve accuracy of span allocation to blocks based on content type

- Update the logic for determining `end_page_id` to handle negative values
- This change ensures proper behavior when `end_page_id` is set to -1 or other negative values

- Add license ID information to the log for better traceability
- Improve logging format to include both license ID and expiration date

- Add ImportError handling to silence known import-related exceptions
- Improve generic exception handling to log error messages- Maintain existing specific exception handlers for license-related issues

- Add license verification logic for Ascend plugin
- Handle different license-related exceptions with appropriate error messages
- Log success message with license expiration date if verification passes
- Fall back to CPU model if license verification fails or plugin is not available

- Update instructions for AI-generated titles optimization
- Use ast.literal_eval() instead of json.loads() for parsing completion content
- Refactor variable names and logging for better code readability- Add error handling for JSON decoding issues

- Move environment variable settings for NPU, MPS, and other configurations to the global scope in doc_analyze_by_custom_model.py
- Remove redundant environment variable settings in pdf_extract_kit.py
- This change ensures consistent configuration across the application and avoids potential conflicts or duplicate settings

- Remove redundant imports for StructTableModel and TableMasterPaddleModel
- Reorder imports to group related modules together
- Update import structure for better readability and maintainability

- Remove unused utility functions
- Update import statements for better readability
- Add conditional imports for Ascend plugin
- Refactor table model initialization to support NPU

- Update calculate_contrast function to support both RGB and BGR image modes
- Add input validation for image mode in calculate_contrast function
- Modify usage of calculate_contrast function in OCR processing to specify image mode

- Increase batch size from 8 to 256 for language detection inference
- Add timing measurement for language detection process

- Rename empty_spans to need_ocr_spans for better clarity
- Add calculate_contrast function to measure image contrast
- Filter out low-contrast spans to improve OCR accuracy
- Update OCR processing workflow to use new filtering method

- Uncomment detect_invalid_chars_by_pymupdf function call
- Comment out detect_invalid_chars function call

- Update batch ratio calculation logic to better utilize available GPU memory
- Improve logging for all GPU memory sizes

- Update batch ratio calculation for GPU memory range
- Increase upper limit for batch ratio 16 from 24 to 32 GB

- Add timing measurement for formula, text, and title optimization using LLM
- Log the execution time for each LLM aided process

- Add a check to return 0 when either bbox1_area or bbox2_area is zero
- This prevents division by zero errors when calculating IoU

- Restore commented code for filtering out characters with invalid bounding boxes
- This change may affect the filtering of unnecessary characters in PDF parsing

- Update conditions for batch ratio assignment:
  -8 <= gpu_memory < 10: batch_ratio = 2 - 10 <= gpu_memory <= 12: batch_ratio =4
- This fix ensures proper batch ratio selection for GPU memory sizes

- Improve batch ratio calculation based on GPU memory
- Enhance performance for devices with 8GB or more VRAM

- Reduce batch_ratio by 1 for better performance and stability
- This change ensures more consistent memory usage when processing documents

refactor(magic_pdf): adjust VRAM allocation and MFR batch size- Update VRAM allocation logic to use 'VIRTUAL_VRAM_SIZE' environment variable
- Reduce MFR (Math Formula Recognition) batch size from 64 to 32

- Update GPU memory check and batch ratio calculation logic
- Add support for virtual VRAM size environment variable
- Improve logging for GPU memory and batch ratio

- Reduce YOLO_LAYOUT_BASE_BATCH_SIZE from 4 to 1
- Simplify batch ratio calculation for formula detection
- Remove unused conditional logic in batch ratio determination