Trying to get the CNNTransformer to work, but it is hard.

3 files changed, 2 insertions, 1 deletions

diff --git a/src/text_recognizer/networks/transformer/positional_encoding.py b/src/text_recognizer/networks/transformer/positional_encoding.py
index a47141b..1ba5537 100644
--- a/src/text_recognizer/networks/transformer/positional_encoding.py
+++ b/src/text_recognizer/networks/transformer/positional_encoding.py
@@ -13,6 +13,7 @@ class PositionalEncoding(nn.Module):
     ) -> None:
         super().__init__()
         self.dropout = nn.Dropout(p=dropout_rate)
+        self.max_len = max_len
 
         pe = torch.zeros(max_len, hidden_dim)
         position = torch.arange(0, max_len).unsqueeze(1)
diff --git a/src/text_recognizer/networks/transformer/sparse_transformer.py b/src/text_recognizer/networks/transformer/sparse_transformer.py
deleted file mode 100644
index 8c391c8..0000000
--- a/src/text_recognizer/networks/transformer/sparse_transformer.py
+++ /dev/null
@@ -1 +0,0 @@
-"""Encoder and Decoder modules using spares activations."""
diff --git a/src/text_recognizer/networks/transformer/transformer.py b/src/text_recognizer/networks/transformer/transformer.py
index 1c9c7dd..c6e943e 100644
--- a/src/text_recognizer/networks/transformer/transformer.py
+++ b/src/text_recognizer/networks/transformer/transformer.py
@@ -230,6 +230,7 @@ class Transformer(nn.Module):
     ) -> Tensor:
         """Forward pass through the transformer."""
         if src.shape[0] != trg.shape[0]:
+            print(trg.shape)
             raise RuntimeError("The batch size of the src and trg must be the same.")
         if src.shape[2] != trg.shape[2]:
             raise RuntimeError(