gradboard 2.3.1__py3-none-any.whl → 2.3.3__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.

This version of gradboard might be problematic. Click here for more details.

Files changed (6) hide show
  1. gradboard/optimiser.py +3 -2
  2. {gradboard-2.3.1.dist-info → gradboard-2.3.3.dist-info}/METADATA +1 -1
  3. gradboard-2.3.3.dist-info/RECORD +8 -0
  4. gradboard-2.3.1.dist-info/RECORD +0 -8
  5. {gradboard-2.3.1.dist-info → gradboard-2.3.3.dist-info}/LICENSE +0 -0
  6. {gradboard-2.3.1.dist-info → gradboard-2.3.3.dist-info}/WHEEL +0 -0
gradboard/optimiser.py CHANGED
@@ -136,13 +136,14 @@ def get_optimiser(model, optimiser=AdamW, lr=1e-3, weight_decay=1e-2):
136
136
  https://github.com/SHI-Labs/Compact-Transformers/blob/main/configs/
137
137
  pretrained/cct_7-3x1_cifar100_1500epochs.yml
138
138
  """
139
+ # TODO: print a warning when params are excluded from weight decay IFF wd is set
139
140
  weight_decay_exclude = []
140
141
  for keyword in [
142
+ "nondecay",
141
143
  "bias",
142
144
  "norm",
143
145
  "embedding",
144
- "swiglu_beta",
145
- "sigma_reparam",
146
+ "beta",
146
147
  "input_query",
147
148
  "reentrant_query",
148
149
  ]:
{gradboard-2.3.1.dist-info → gradboard-2.3.3.dist-info}/METADATA RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.3
2
2
  Name: gradboard
3
- Version: 2.3.1
3
+ Version: 2.3.3
4
4
  Summary: Easily snowboard down gnarly loss gradients
5
5
  License: MIT
6
6
  Author: Nicholas Bailey
gradboard-2.3.3.dist-info/RECORD ADDED
@@ -0,0 +1,8 @@
1
+ gradboard/__init__.py,sha256=57AkHusYwLCsusiVnajH5pMFKioRCj-3IjF9qpdOzE0,69
2
+ gradboard/cycles.py,sha256=iGEW3Rlp-JNyQZLfpaDXxUCAcEV01ANjLF-Fnhug-qA,10120
3
+ gradboard/optimiser.py,sha256=OPWaFSludziX5083NolZOg9BclQFjDbzNz5zsxbERtU,6265
4
+ gradboard/scheduler.py,sha256=TlfdIGx5ufb5dXrhWbDkpePfqLMoHJSGZbdauPgd2FU,6770
5
+ gradboard-2.3.3.dist-info/LICENSE,sha256=0BAzJE5BqQ7Iixp_AFdB2W1uO-HCRX-Qfun8PHt6yVM,1073
6
+ gradboard-2.3.3.dist-info/METADATA,sha256=MSn2XRpfiSRzOCkHcH_FYZate5ONKF1PM8GeJYhJ4XI,2173
7
+ gradboard-2.3.3.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
8
+ gradboard-2.3.3.dist-info/RECORD,,
gradboard-2.3.1.dist-info/RECORD DELETED
@@ -1,8 +0,0 @@
1
- gradboard/__init__.py,sha256=57AkHusYwLCsusiVnajH5pMFKioRCj-3IjF9qpdOzE0,69
2
- gradboard/cycles.py,sha256=iGEW3Rlp-JNyQZLfpaDXxUCAcEV01ANjLF-Fnhug-qA,10120
3
- gradboard/optimiser.py,sha256=IDlF0uLwubq9gUQJx25aNiSeBYoBfxFRKWMm5OkiZzU,6192
4
- gradboard/scheduler.py,sha256=TlfdIGx5ufb5dXrhWbDkpePfqLMoHJSGZbdauPgd2FU,6770
5
- gradboard-2.3.1.dist-info/LICENSE,sha256=0BAzJE5BqQ7Iixp_AFdB2W1uO-HCRX-Qfun8PHt6yVM,1073
6
- gradboard-2.3.1.dist-info/METADATA,sha256=WoHkaWSqSxfauQRsn3qWyWTurFEQkg6JBSCIUgJU49c,2173
7
- gradboard-2.3.1.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
8
- gradboard-2.3.1.dist-info/RECORD,,