Merge pull request #36 from LukasHedegaard/develop

Add `nn.LayerNorm` to automatically convertible modules

Author: LukasHedegaard

.gitignore

@@ -129,3 +129,5 @@ dmypy.json
 .pyre/
 
 .vscode
+
+.DS_Store

CHANGELOG.md

@@ -9,6 +9,10 @@ From v1.0.0 and on, the project will adherence strictly to Semantic Versioning.
 
 ## [Unreleased]
 
+## [0.15.1]
+### Added
+- `nn.LayerNorm` to automatically convertible modules
+
 
 ## [0.15.0]
 ### Added
@@ -19,6 +23,9 @@ From v1.0.0 and on, the project will adherence strictly to Semantic Versioning.
 - Sequential `padding` computation.
 - `Lambda` `__repr__` function prints.
 
+### Removed
+- CI testing for python v3.6.
+
 
 ## [0.14.0]
 ### Added

README.md

@@ -266,10 +266,11 @@ In addition, we support interoperability with a wide range of modules from `torc
     - `nn.Softmax2d`
     - `nn.LogSoftmax`
 
-- Batch Normalisation
+- Normalisation
     - `nn.BatchNorm1d`
     - `nn.BatchNorm2d`
     - `nn.BatchNorm3d`
+    - `nn.LayerNorm`
 
 - Dropout
     - `nn.Dropout`

continual/convert.py

@@ -136,6 +136,7 @@ def forward_with_callmode(*args, **kwargs):
     nn.BatchNorm1d,
     nn.BatchNorm2d,
     nn.BatchNorm3d,
+    nn.LayerNorm,
     # >> Dropout modules
     nn.Dropout,
     nn.Dropout2d,

setup.py

@@ -25,7 +25,7 @@ def from_file(file_name: str = "requirements.txt", comment_char: str = "#"):
 
 setup(
     name="continual-inference",
-    version="0.15.0",
+    version="0.15.1",
     description="Building blocks for Continual Inference Networks in PyTorch",
     long_description=long_description(),
     long_description_content_type="text/markdown",