Skip to content

CHANGELOG.md

Latest commit

 

History

History
94 lines (79 loc) · 4.36 KB

CHANGELOG.md

File metadata and controls

94 lines (79 loc) · 4.36 KB

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.

[0.1.1] - 2025-01-20

Added

  • GPU Support: Complete GPU acceleration for GNN training with automatic device detection (CUDA → MPS → CPU fallback)
  • Device Utilities: New device_utils.py module with intelligent device management and optimization tips
  • Performance Enhancements: Memory usage monitoring and device-specific optimization settings
  • Generic Model Comparison: New model_comparison.py module for comparing ontology vs vanilla models
  • Improved Testing: GPU support validation and end-to-end training tests

Fixed

  • Broken Compare Command: Fixed on2vec compare command that was importing non-existent test_edam_model.py
  • Device Compatibility: All models, training, and embedding generation now support GPU acceleration
  • Import Issues: Resolved broken CLI imports and dependencies
  • Tensor Device Mismatches: Fixed tensor placement issues in multi-device scenarios

Changed

  • Enhanced Training Pipeline: All training workflows now support device specification and GPU acceleration
  • Better Error Handling: Improved device detection with fallback strategies and informative warnings
  • Memory Management: Added GPU memory monitoring and optimization recommendations
  • Model Architecture: Updated all GNN models (GCN, GAT, RGCN, Heterogeneous) to be device-aware

Performance

  • Significant Speed Improvements: Text-augmented training shows major performance gains on GPU (e.g., 0.08s on MPS)
  • Memory Efficiency: Optimized tensor operations for GPU memory usage
  • Automatic Optimization: Device-specific settings for optimal performance

Technical Improvements

  • Added comprehensive device detection and validation
  • Enhanced error messages with device context
  • Improved model checkpointing with device state preservation
  • Better cross-platform compatibility (CUDA, MPS, CPU)

Documentation

  • Added GPU setup and usage documentation
  • Performance benchmarking results and recommendations
  • Device-specific optimization guides

[0.1.0] - 2025-01-18

Added

  • Initial release of on2vec toolkit
  • Unified CLI with on2vec command and subcommands
  • Core GNN training workflows (train, embed)
  • HuggingFace Sentence Transformers integration (hf, hf-train, hf-create, hf-test)
  • MTEB benchmarking support (benchmark, compare)
  • Multiple GNN architectures: GCN, GAT, RGCN, Heterogeneous
  • Text-augmented training with sentence transformers
  • Multiple fusion methods for combining structural and text embeddings
  • Auto-generated model cards with comprehensive metadata
  • Batch processing capabilities for multiple ontologies
  • Comprehensive test suite with 30+ tests
  • Rich CLI with help, error handling, and progress indicators

Features

  • Graph Neural Networks: Support for GCN, GAT, RGCN, and heterogeneous models
  • Text Integration: Combines structural graph embeddings with text features
  • HuggingFace Compatibility: Creates drop-in replacement sentence transformer models
  • MTEB Benchmarking: Full integration with Massive Text Embedding Benchmark
  • Smart Automation: Auto-detects base models, domains, and configurations
  • Professional Output: Auto-generated model cards, upload instructions, and documentation
  • Batch Processing: Efficient processing of multiple ontologies
  • Multiple Loss Functions: Triplet, contrastive, cosine, cross-entropy
  • Visualization: UMAP-based embedding visualization
  • Cross-Format Support: Parquet, CSV conversion utilities

Architecture

  • Modular Python package structure
  • CLI-first design with unified entry point
  • Comprehensive error handling and logging
  • Test-driven development with pytest
  • Clean separation between core functionality and CLI

Dependencies

  • Python >= 3.10
  • PyTorch + torch-geometric for GNN training
  • owlready2 for OWL ontology processing
  • sentence-transformers for text features
  • polars for efficient data handling
  • MTEB for benchmarking (optional)
  • HuggingFace Hub for model sharing

Documentation

  • Complete README with quickstart guide
  • CLI reference documentation
  • HuggingFace integration guide
  • Comprehensive docstrings
  • Usage examples and tutorials