(Please add entries here with your changes for the next version)
- ...
- Added Reinforcement Learning example (examples/ppo).
- Fix Adafactor bug that prevented factorization.
- Fix scan broadcast issue in functional core.
- Fix initialization RNGs to work with omnistaging for jitted inits.
- Replaces usage of 'param' kind to 'params' collection.
- Fix LARS optimizer for zero param initialization.
- Added various examples in Linen API. See README.md for more information.
- Full JAX omnistaging compatibility.
- Added JAX trace-level checks for transforms.
- BatchNorm added axis_index_groups for control in parallel training.
- Optimizers broken out into separate directory with base class and implementations.
- traverse_util added flatten_dict and unflatten_dict utility methods for nested dicts.
- Add ConvTranspose Module to nn.linear
- Rename the following optional arguments to nn.linear.Conv:
lhs_dilation->input_dilation,rhs_dilation->kernel_dilation - Change default layer names from numbers '0', '1', etc. to include the Module class name, e.g. 'Dense_0', 'LayerNorm_1'.