broccoli-ml 0.39.0__py3-none-any.whl → 0.40.0__py3-none-any.whl

broccoli/vit.py

@@ -120,6 +120,7 @@ class ViTEncoder(nn.Module):
         transformer_initial_ff_residual_path=True,
         transformer_initial_ff_linear_module_up=None,
         transformer_initial_ff_linear_module_down=None,
+        transformer_initial_ff_mlp_dropout=None,
         transformer_pre_norm=True,
         transformer_normformer=False,
         transformer_post_norm=False,
@@ -307,7 +308,12 @@ class ViTEncoder(nn.Module):
                 transformer_embedding_size,
                 activation=transformer_activation,
                 activation_kwargs=transformer_activation_kwargs,
-                dropout=transformer_mlp_dropout,
+                dropout=(
+                    # First truthy assigned value
+                    transformer_initial_ff_mlp_dropout
+                    if transformer_initial_ff_mlp_dropout is not None
+                    else transformer_mlp_dropout
+                ),
                 linear_module_up=(
                     # First truthy assigned value
                     transformer_initial_ff_linear_module_up
@@ -349,11 +355,7 @@ class ViTEncoder(nn.Module):
 
 class ViT(nn.Module):
     """
-    Denoising convolutional transformer
-    Based on the Compact Convolutional Transformer (CCT) of [Hasani et al. (2021)
-        *''Escaping the Big Data Paradigm with Compact Transformers''*](
-        https://arxiv.org/abs/2104.05704). It's a convolutional neural network
-        leading into a transformer encoder, followed by a sequence pooling layer.
+    ...
     """
 
     def __init__(
@@ -380,6 +382,7 @@ class ViT(nn.Module):
         transformer_initial_ff_residual_path=True,
         transformer_initial_ff_linear_module_up=None,
         transformer_initial_ff_linear_module_down=None,
+        transformer_initial_ff_mlp_dropout=None,
         transformer_pre_norm=True,
         transformer_normformer=False,
         transformer_post_norm=False,
@@ -442,6 +445,7 @@ class ViT(nn.Module):
             transformer_initial_ff_residual_path=transformer_initial_ff_residual_path,
             transformer_initial_ff_linear_module_up=transformer_initial_ff_linear_module_up,
             transformer_initial_ff_linear_module_down=transformer_initial_ff_linear_module_down,
+            transformer_initial_ff_mlp_dropout=transformer_initial_ff_mlp_dropout,
             transformer_pre_norm=transformer_pre_norm,
             transformer_normformer=transformer_normformer,
             transformer_post_norm=transformer_post_norm,

{broccoli_ml-0.39.0.dist-info → broccoli_ml-0.40.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: broccoli-ml
-Version: 0.39.0
+Version: 0.40.0
 Summary: Some useful Pytorch models, circa 2025
 License: MIT
 Author: Nicholas Bailey

{broccoli_ml-0.39.0.dist-info → broccoli_ml-0.40.0.dist-info}/RECORD

@@ -10,8 +10,8 @@ broccoli/rope.py,sha256=hw7kBPNR9GQXj4GxyIAffsGKPfcTPOFh8Bc7oEHtaZY,12108
 broccoli/tensor.py,sha256=um8mrxkYbvNDo-QvHlmJm8Aw6qcngOlUZPoAk_PMReA,4480
 broccoli/transformer.py,sha256=Xw1oBLsvVeHmMqgurhorRa49nrjfooLev5uBPFeK9og,17004
 broccoli/utils.py,sha256=htq_hOsdhUhL0nJi9WkKiEYOjEoWqFpK5X49PtgTf-0,299
-broccoli/vit.py,sha256=RaSJZh2ogqNAvAkQDuZpNqdtRTWaW_8ug4BsBCBK_f4,17728
-broccoli_ml-0.39.0.dist-info/LICENSE,sha256=0BAzJE5BqQ7Iixp_AFdB2W1uO-HCRX-Qfun8PHt6yVM,1073
-broccoli_ml-0.39.0.dist-info/METADATA,sha256=wR3glIOttDNALtYadYXNBbwjHS3p5wtBoWGwjLEg8c8,1257
-broccoli_ml-0.39.0.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-broccoli_ml-0.39.0.dist-info/RECORD,,
+broccoli/vit.py,sha256=qnA7Utyp83CQ2o7kdkDB0d9Yj2UHFOVhNF8TtqwefIc,17781
+broccoli_ml-0.40.0.dist-info/LICENSE,sha256=0BAzJE5BqQ7Iixp_AFdB2W1uO-HCRX-Qfun8PHt6yVM,1073
+broccoli_ml-0.40.0.dist-info/METADATA,sha256=f9pHDub9pIvrB5SIoiTsfAKrHRhHyZGFNdWNJhDwnH0,1257
+broccoli_ml-0.40.0.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+broccoli_ml-0.40.0.dist-info/RECORD,,