Continual-inference

Latest version: v1.2.3

Safety actively analyzes 681874 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 8 of 8

0.2.2

Added
- Automatic conversion of batch normalization and activation functions.

Fixed
- Separate dilation and stride in pool.

Changed
- Conv forward to use temporal padding like (like nn.Conv).

Removed
- `co.BatchNorm2d`

0.2.1

Changed
- Renamed `unsqueezed` to `forward_stepping`.

Removed
- Unused utility `Zeros`

0.2.0

Changed
- Naming to match `torch.nn`. This lets the continual modules be used as drop-in replacements for `torch.nn` modules.
- Renamed `forward_regular_unrolled` to `forward`, `forward_regular` to `forward_steps`, and `forward` for `forward_step`.
- Renamed `from_regular` to `build_from`.
- Renamed `continual` to `unsqueezed`.

Added
- `Sequential` wrapper for sequential application of modules
- `Parallel` wrapper for parallel application and aggregation of inputs
- `Residual` wrapper for adding a unity residual to a module
- `continual` conversion function
- `register` function for 3rd party modules to register their conversion
- Additional tests

0.1.2

Added
- Pooling modules: `MaxPool1d`, `AvgPool3d`, `MaxPool3d`, `AdaptiveAvgPool3d`, `AdaptiveMaxPool3d`.

0.1.1

Added
- Updated README.

0.1.0

Added
- Initial publicly available implementation of the library.

Page 8 of 8

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.