Spaces:
Sleeping
Sleeping
Commit History
-
Add name translation fix for OPUS-MT - correct 'sand geese' to 'Zhang Shaian'
Peter Yang- Add _fix_name_translations() method to post-process OPUS-MT translations - Fix incorrect translation of 章沙雁 (Zhang Shaian) as 'sand geese' - Handle 'elders of the sand geese' -> 'Zhang Shaian Elder' - Preserve 'Elder' title when original text contains 长老 - Apply fix automatically after OPUS-MT translatione632f5c -
Revert to OPUS-MT translation by default - better name handling
Peter Yang- Change default use_qwen_translation to False in DocumentProcessingAgent - Change default use_qwen_translation to False in WorshipProgramGenerator - Update app.py to explicitly use OPUS-MT (False) - OPUS-MT correctly translates names like 章沙雁 → Zhang Shaian - Qwen was incorrectly translating names (e.g., 'sand gee' instead of proper name)f04ca50
-
Fix translation hallucinations by adding stop sequences and improving cleanup
Peter Yang- Add StoppingCriteria to prevent model from continuing generation - Implement custom StopSequenceCriteria class for Qwen chat format - Move hallucination detection earlier (before cleanup steps) - Simplify cleanup logic from 6 layers to 3 focused steps - Improve extraction logic to handle Qwen chat format better - Add comprehensive code review documentation44c353a -
Improve translation prompt to ensure completeness for titles and all sentences
Peter Yang531486f -
Improve translation completeness: better paragraph splitting for headings, improved prompts, lower min length for titles
Peter Yangd3816fa -
Fix: Enable PDF processing for content extraction and improve short translation handling
Peter Yang9124732 -
Fix Qwen2.5 translation truncation: increase max_length to 1024, dynamic max_new_tokens (300-800), less aggressive cleanup
Peter Yang1c3d743 -
Remove unused messages variable
Peter Yang52fab4c -
Fix: Save bilingual file to current directory, prevent PDF/bilingual file duplication, ensure Message section only uses bilingual content
Peter Yang917635f -
Integrate improved Qwen2.5 translation into document_processing_agent with fallback to OPUS-MT
Peter Yangb68a16e
-
Fix: Use actual bilingual file from document_sources instead of hardcoded test file
Peter Yang6cea639 -
Initial commit for HF Space deployment
Peter Yang3307396