Skip to content

CHANGELOG.md

Latest commit

 

History

History
116 lines (95 loc) · 4.26 KB

CHANGELOG.md

File metadata and controls

116 lines (95 loc) · 4.26 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[2.0.0] - 2026-09-16

Changed

  • Migrated to DeepSeek-OCR-2 (deepseek-ai/DeepSeek-OCR-2, released Jan 2026)
    • MODEL_NAME default updated in config.py, .env.example, and docker-compose.yml
    • IMAGE_SIZE default 640 → 768 (OCR-2 dynamic resolution: (0-6)×768×768 + 1×1024×1024)
  • Runtime upgraded: Python 3.10 → 3.12, torch 2.0.1 → 2.6.0, torchvision 0.21.0 (cu118)
  • sentencepiece 0.1.99 → ≥0.2.0 (cp312 wheel; 0.1.99 fails to build from source)
  • Docker Compose: GPU reservation moved to docker-compose.gpu.yml overlay — docker compose up works on any host, docker compose -f docker-compose.yml -f docker-compose.gpu.yml up enables NVIDIA GPU
  • Backend container pinned to platform: linux/amd64 (PyTorch cu118 wheels are x86_64-only)

Removed

  • test_compress parameter from model.infer() calls (not accepted by the OCR-2 API)

Added

  • Backend architecture refactor (deep modules, testable without GPU):
    • backend/model_lifecycle.py — ModelLifecycle (async-lock load state machine), Progress/ModelStatus/InferSettings Pydantic models, OcrModel/LoadableModel Protocols, DeepSeekModel adapter (owns flash_attention_2 → eager fallback)
    • backend/ocr_handler.py — OcrHandler request pipeline (validation → upload → prompt resolution → inference → result assembly)
    • backend/tests/ — pytest suite with FakeOcr adapter (7 tests, no GPU/model needed)
    • backend/requirements-dev.txt — pytest + pytest-asyncio
  • GitHub Actions release workflow (tag-triggered, GitHub-native release notes)

Fixed

  • Race window in model load (module-global flags replaced by asyncio.Lock around ensure_loaded())
  • main.py reduced from 414 to 216 lines; endpoints are now thin shells

[1.0.0] - 2025-01-26

Added

  • Initial release of DeepSeek OCR web application
  • FastAPI backend with OCR processing endpoints
  • Modern web interface with drag-and-drop file upload
  • Progress bar for model download with real-time updates
  • Demo mode to test interface without downloading model
  • Multiple OCR modes: Free OCR, Markdown, Grounding, Parse Figure, Detailed
  • Docker Compose deployment setup
  • GPU acceleration support (NVIDIA CUDA 11.8+)
  • Health check endpoints
  • API documentation with Swagger/OpenAPI
  • Comprehensive project documentation
  • Test script for API validation

Features

  • Real-time model download progress tracking
  • Manual model download trigger button
  • Background model loading
  • Image preview before processing
  • Result copying and downloading
  • Responsive web design
  • Error handling and user feedback
  • Multiple OCR processing modes with custom prompts

Documentation

  • README with quick start guide
  • LICENSE (MIT - Development and Testing Only)
  • CONTRIBUTING guidelines
  • CODE_OF_CONDUCT
  • SECURITY policy
  • QUICKSTART guide
  • USAGE_GUIDE comprehensive manual
  • WINDOWS_SETUP guide
  • CHANGELOG

Technical

  • Optimized Docker images (using runtime instead of devel)
  • Minimal system dependencies
  • BuildKit cache support for faster rebuilds
  • Proper error handling and logging
  • CORS configuration for API access
  • File upload validation
  • Health monitoring

Security Notes

  • ⚠️ Designed for development and testing only
  • No authentication implemented
  • No rate limiting
  • Open CORS policy
  • Not hardened for production use

[Unreleased]

Planned

  • User authentication and authorization
  • Rate limiting
  • Enhanced security features
  • Multi-language support for UI
  • Batch processing support
  • Result history and management
  • Advanced configuration options
  • Performance optimizations
  • Additional OCR modes
  • PDF multi-page support

Version History

How to Read This Changelog

  • Added: New features
  • Changed: Changes in existing functionality
  • Deprecated: Soon-to-be removed features
  • Removed: Removed features
  • Fixed: Bug fixes
  • Security: Security improvements

Note: This project is for development and testing purposes only. Production use is not recommended and done at your own risk.