apunkt/dp-fusion-lib
1
0
Fork 0
mirror of https://github.com/rushil-thareja/dp-fusion-lib.git synced 2026-04-24 12:06:23 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions
dp-fusion-lib/CHANGELOG.md
rushil-thareja d012046d85 Initial release v0.1.0 
- Token-level differential privacy for LLMs
  - Integration with Document Privacy API
  - Comprehensive test suite and documentation
  - Examples and Jupyter notebook included
2025-12-23 17:02:06 +04:00

1.3 KiB
Raw Permalink Blame History

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.

[Unreleased]

[0.1.0] - 2025-01-01

Added

  • Initial release of DP-Fusion-Lib
  • DPFusion class for differentially private text generation
    • Message-based context building with add_message()
    • Direct context generation with generate()
    • Token-level generation with generate_from_tokens()
  • Tagger class for automatic private phrase extraction
    • Integration with Document Privacy API
    • Support for multiple document types (HEALTH, FINANCE, LEGAL)
  • Privacy accounting functions
    • compute_epsilon_single_group() for single-group privacy guarantees
    • compute_dp_epsilon() for multi-group scenarios
  • Utility functions for advanced usage
    • compute_renyi_divergence_clipped_symmetric() for divergence computation
    • find_lambda() for binary search of mixing parameter
    • replace_sequences_with_placeholder_fast() for token-level redaction
  • Support for HuggingFace transformers models
  • Incremental decoding with KV-cache optimization
  • Comprehensive documentation and examples

Dependencies

  • PyTorch >= 2.0.0
  • Transformers >= 4.25.0
  • Requests >= 2.25.0