Roadmap

Roadmap#

This document outlines potential future directions for Panther development. These are aspirational goals and not commitments to specific timelines or features.

Current Status (v0.1.2)#

Available Features

  • ✅ Core sketching algorithms (CQRRPT, Randomized SVD)

  • ✅ SKLinear layer implementation

  • ✅ SKConv2d layer implementation

  • ✅ RandMultiHeadAttention mechanism

  • ✅ CUDA kernel support

  • ✅ PyTorch integration

  • ✅ C++/CUDA backend (pawX)

  • ✅ SKAutoTuner with multiple search algorithms

  • ✅ Documentation and examples

Current Limitations

  • Single-node training only

  • Limited to implemented layer types

  • CUDA 12.4+ required for GPU acceleration (CPU-only mode fully supported)

  • Windows and Linux fully supported (macOS experimental, CPU-only)

Potential Future Directions#

The following are areas being considered for future development. Community contributions and feedback will help shape priorities.

Enhanced Neural Network Support

  • Additional sketched layer types (LSTM, GRU, etc.)

  • More attention mechanism variants

  • Additional convolution operation types

  • Transformer block optimizations

Performance Optimizations

  • Enhanced CUDA kernel implementations

  • Dynamic sketching with adaptive parameters

  • Support for additional GPU architectures

Platform and Framework Support

  • Improved macOS support

  • ARM64 architecture support

  • Potential integration with other frameworks

  • Multi-GPU training support

Developer Tools

  • Enhanced debugging and profiling utilities

  • Additional AutoTuner optimization strategies

  • Better performance analysis tools

  • Expanded documentation and examples

Algorithm Improvements

  • Additional sketching methods

  • Hierarchical sketching approaches

  • Task-specific optimization strategies

  • Improved accuracy-performance tradeoffs

Contributing#

Community contributions are welcome! If you’re interested in working on any of these areas or have other ideas, please:

  • Open an issue on GitHub to discuss your proposal

  • Check existing issues for areas where help is needed

  • Submit pull requests with improvements

The actual development path will depend on:

  • Community needs and feedback

  • Contributor interests and availability

  • Technical feasibility

  • Resource availability

For questions or suggestions about future directions, please open an issue on the GitHub repository.