Known Issues

Known Issues (Summary)

This file lists only unique, actionable issues not covered in other documentation. For installation, usage, contributing, and roadmap, see the respective guides.

My known issues

  • Import time is long in the Example_Attack_Notebook.ipynb ~40s

  • Generally still have high overhead +1000%.

  • CUDA not tested yet

AI-Identified Issues

1. Examples Directory - Pandas Import Performance

Some example scripts (e.g. basic_cnn_robustness.py) may hang during pandas import in certain Python environments. Use uv run python for faster startup. Does not affect framework usage as a library.

2. Performance Optimization

Some functions (e.g. bitflip_float32) are not fully vectorized. Optimized versions exist but further improvements are possible, especially for GPU workloads.

3. Error Handling

Exception types are not fully standardized. Custom exception classes are planned for future releases.

4. Test Coverage

Some edge cases (NaN/Inf, multi-GPU, custom criterion) are not fully tested. Coverage is high (>90%) but not exhaustive.

5. Unused Code

Some utility functions (e.g. get_model_info in device.py) appear unused and may be removed in future cleanups.

For details on planned improvements, contributing, or reporting issues, see:

  • CONTRIBUTING.md

  • CHANGELOG.md

  • docs/installation.md

  • docs/quickstart.md

For the full TODO system and categories, refer to code comments and the contributing guide.