broccoli-ml 0.9.0__tar.gz → 0.11.0__tar.gz
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.
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/PKG-INFO +1 -1
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/transformer.py +4 -5
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/vit.py +0 -1
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/pyproject.toml +1 -1
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/LICENSE +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/README.md +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/__init__.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/activation.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/assets/2025_resnet_imagenet_1k_pretrained_state_dict.pkl +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/assets/cifar100_eigenvectors_size_2.pt +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/assets/cifar100_eigenvectors_size_3.pt +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/cnn.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/eigenpatches.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/linear.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/rope.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/tensor.py +0 -0
- {broccoli_ml-0.9.0 → broccoli_ml-0.11.0}/broccoli/utils.py +0 -0
@@ -233,7 +233,6 @@ class FeedforwardLayer(nn.Module):
|
|
233
233
|
output_features,
|
234
234
|
activation=nn.ReLU,
|
235
235
|
activation_kwargs=None,
|
236
|
-
layer_norm=True,
|
237
236
|
dropout=0.0,
|
238
237
|
linear_module=nn.Linear,
|
239
238
|
):
|
@@ -248,17 +247,17 @@ class FeedforwardLayer(nn.Module):
|
|
248
247
|
|
249
248
|
self.process = nn.Sequential(
|
250
249
|
*[
|
251
|
-
nn.LayerNorm(input_features)
|
250
|
+
nn.LayerNorm(input_features),
|
252
251
|
linear_module(
|
253
252
|
input_features,
|
254
253
|
(
|
255
|
-
2 * ratio *
|
254
|
+
2 * ratio * output_features
|
256
255
|
if activation.__name__.endswith("GLU")
|
257
|
-
else ratio *
|
256
|
+
else ratio * output_features
|
258
257
|
),
|
259
258
|
),
|
260
259
|
self.activation,
|
261
|
-
linear_module(ratio *
|
260
|
+
linear_module(ratio * output_features, output_features),
|
262
261
|
self.dropout,
|
263
262
|
]
|
264
263
|
)
|
@@ -261,7 +261,6 @@ class ViTEncoder(nn.Module):
|
|
261
261
|
transformer_embedding_size,
|
262
262
|
activation=transformer_activation,
|
263
263
|
activation_kwargs=transformer_activation_kwargs,
|
264
|
-
layer_norm=False, # rely on initial batch_norm
|
265
264
|
dropout=transformer_mlp_dropout,
|
266
265
|
linear_module=linear_module,
|
267
266
|
)
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|