Releases: metatensor/metatensor
metatensor-core v0.1.11
This is a patch release of metatensor-core, mainly improving performance of Labels
creation when handling large set of labels.
metatensor-operations v0.2.4
This is a small patch release, adding the ability to give names to the labels generated in block_from_array
(added in #746)
metatensor-torch v0.5.5
metatensor-torch v0.5.4
This release brings the new Labels.select
and TensorBlock
serialization to the TorchScript backend.
It also adds functionalities to store and retrieve custom metadata inside saved atomistic models.
Finally, this release re-enables pre-built wheels on Windows compatible with Torch v2.3 and v2.4.
metatensor-operations v0.2.3
This is a small patch release with the following changes:
- We now require Python >= 3.9
slice
anddrop_blocks
are now faster thanks toLabels.select
metatensor-learn v0.2.3
This is a new patch release of metatensor-learn, the main addition is the ability to have fields which are not valid Python identifiers in Dataset
and DataLoader
.
metatensor-core v0.1.10
This release brings two new functionalities to metatensor-core:
Labels.select
, a way to find label entries matching a selectionTensorBlock
serialization, enabling the saving & loading of standaloneTensorBlock
metatensor-torch v0.5.3
This is a patch release of metatensor-torch, with the following changes
Changed
MetatensorAtomisticModel.save()
always saves models on the CPU.- We now require Python >= 3.9
Fixed
- Fixed a memory leak in
register_autograd_neighbors
(#684)
metatensor-core v0.1.9
This is a patch release of metatensor-core, with the following changes:
Added
TensorBlock.__len__
andTensorBlock.shape
, which return the length and shape of the valu es in the block respectively- We can now load (but not save) TensorMap stored in npz files using DEFLATE compression (#671)
Changed
- We now require Python >= 3.9
Fixed
- Fixed a memory leak affecting all data stored in TensorBlock (#683)
metatensor-torch v0.5.2
This is a patch release of metatensor-torch, containing the following changes:
Added
MetatensorAtomisticModel.save()
to save a wrapped model to a file.TensorBlock.__len__
andTensorBlock.shape
, which return the length and shape of the values in the block respectively (#640)metatensor.torch.atomistic.ase_calculator.MetatensorCalculator
can now usevesin
for faster neighbor list calculations (#659)- When running atomistic models in the PyTorch profiler, different sections of the code now have meaningful names
Deprecated
MetatensorAtomisticModel.export()
is deprecated in favor ofMetatensorAtomisticModel.save()
Fixed
metatensor.torch.atomistic.ase_calculator.MetatensorCalculator
uses the right device when computing stress/virial (#660)