gradboard 2.3.0__tar.gz → 2.3.2__tar.gz

{gradboard-2.3.0 → gradboard-2.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: gradboard
-Version: 2.3.0
+Version: 2.3.2
 Summary: Easily snowboard down gnarly loss gradients
 License: MIT
 Author: Nicholas Bailey

{gradboard-2.3.0 → gradboard-2.3.2}/gradboard/optimiser.py

@@ -136,13 +136,14 @@ def get_optimiser(model, optimiser=AdamW, lr=1e-3, weight_decay=1e-2):
         https://github.com/SHI-Labs/Compact-Transformers/blob/main/configs/
         pretrained/cct_7-3x1_cifar100_1500epochs.yml
     """
+    # TODO: print a warning when params are excluded from weight decay IFF wd is set
     weight_decay_exclude = []
     for keyword in [
         "bias",
         "norm",
         "embedding",
-        "swiglu_beta",
-        "sigma_reparam" "input_query",
+        "beta",
+        "input_query",
         "reentrant_query",
     ]:
         weight_decay_exclude += [

{gradboard-2.3.0 → gradboard-2.3.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "gradboard"
-version = "2.3.0"
+version = "2.3.2"
 description = "Easily snowboard down gnarly loss gradients"
 authors = [
     {name = "Nicholas Bailey"}