Releases: google-research/dataclass_array
Releases · google-research/dataclass_array
v1.5.2
v1.5.1
- Fix look_at compatibility with pytorch
- Add
dca.vectorize_methodcompatibility fortf.nest/tf.data.
v1.5.0
- Add
tf.nest/tf.datacompatibility forDataclassArray. - Fix import error due to
etils
v1.4.2
- Add
dca.concatmethod in addition todca.stack. - Now require Python 3.9 (drop 3.8 support)
v1.4.1
- Add
torch==2.0.0support
v1.4.0
- Add
torchsupport! - Add
.cpu(),.cuda(),.to()methods to move the dataclass from
devices when using torch. - Breaking:
@dataclass(frozen=True)is now automatically applied
v1.3.0
- Added: Support for static
dca.DataclassArray(dataclasses with only
static fields).
v1.2.1
- Fixed: Compatibility with
edc.dataclass(auto_cast=True)(fix the'type' object is not subscriptableerror)
v1.2.0
- Changed: By default, dataclass_array do not cast and broadcast inputs
anymore. - Changed:
dca.DataclassArrayfields can be annotated with named axis (e.g.
FloatArray['*shape h w 3']). Note that consistency across fields is not
checked yet. - Added:
@dca.dataclass_arrayto customize thedca.DataclassArrayparams
v1.1.0
- Added: Array types can be imported directly from
dataclass_array.typing - Added: Syntax to specify the shape of the DataclassArray (e.g.
MyRay['h w']). - Fixed: Correctly forward non-init fields in
.replace,tree_map,
@dca.vectorize_method
Previous Next