v2.12.1

Compare Source

Fixed

• tensorizer is now compatible with Python 3.14

v2.12.0

Compare Source

Added

• Tensors with long dimensions (≥ 2 ** 32 elements in a single dimension)
  can now be serialized and deserialized

Changed

• tensorizer.utils.CPUMemoryUsage.free now reports available memory
  rather than free memory

Fixed

• tensorizer.torch_compat can now serialize and deserialize tensors that have
  storages with sizes ≥ 2 ** 32

v2.11.1

Compare Source

Fixed

• Fixed an incompatibility between tensorizer.torch_compat and torch<2.7.1
  • Thank you to stuart-mv for catching this

v2.11.0

Compare Source

Added

• tensorizer.torch_compat is a new module for using tensorizer as a backend
  for handling tensor data during standard torch.save and torch.load calls
  • To use tensorizer as a backend for torch.save,
    wrap the call in the tensorizer_saving context manager
    • The file created must then be loaded using tensorizer_loading
  • To use tensorizer as a backend for torch.load,
    wrap the call in the tensorizer_loading context manager
    • The file to load must have been created using tensorizer_saving