remove default, update docstring and test case

i6_models/parts/conformer/convolution.py

@@ -99,10 +99,9 @@ class ConformerConvolutionV2Config(ConformerConvolutionV1Config):
     New attribute:
         dropout_broadcast_axes: string of axes to which dropout is broadcast, e.g. "T" for broadcasting to the time axis
                                 setting to None to disable broadcasting
-    Allows even kernel size
     """
 
-    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]] = None
+    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]]
 
     def check_valid(self):
         assert self.kernel_size % 2 == 1, "ConformerConvolutionV1 only supports odd kernel sizes"

i6_models/parts/conformer/feedforward.py

@@ -62,14 +62,19 @@ def forward(self, tensor: torch.Tensor) -> torch.Tensor:
 
 
 @dataclass
-class ConformerPositionwiseFeedForwardV2Config(ConformerPositionwiseFeedForwardV1Config):
+class ConformerPositionwiseFeedForwardV2Config(ModelConfiguration):
     """
     New attribute:
         dropout_broadcast_axes: string of axes to which dropout is broadcast, e.g. "T" for broadcasting to the time axis
                                 setting to None to disable broadcasting
+    Default value for `activation` removed
     """
 
-    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]] = None
+    input_dim: int
+    hidden_dim: int
+    dropout: float
+    activation: Callable[[torch.Tensor], torch.Tensor]
+    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]]
 
     def check_valid(self):
         assert self.dropout_broadcast_axes in [

i6_models/parts/conformer/mhsa.py

@@ -73,7 +73,7 @@ class ConformerMHSAV2Config(ConformerMHSAV1Config):
                                 setting to None to disable broadcasting
     """
 
-    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]] = None
+    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]]
 
     def check_valid(self):
         assert self.dropout_broadcast_axes in [

i6_models/parts/conformer/mhsa_rel_pos.py

@@ -45,7 +45,7 @@ class ConformerMHSARelPosV1Config(ModelConfiguration):
     separate_pos_emb_per_head: bool
     pos_emb_dropout: float
     dropout: float
-    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]] = None
+    dropout_broadcast_axes: Optional[Literal["B", "T", "BT"]]
 
     def __post_init__(self) -> None:
         super().__post_init__()

requirements_dev.txt

@@ -1,4 +1,4 @@
 onnx
 onnxruntime
 espnet
-typeguard
+typeguard >= 4.3.0

tests/test_conformer_rel_pos.py

@@ -140,6 +140,8 @@ def test_ConformerMHSARelPosV1_against_Espnet():
     dropout_rate = 0.1
     batch_dim_size = 4
     time_dim_size = 50
+    seq_len = torch.Tensor([50, 10, 20, 40])
+    sequence_mask = torch.less(torch.arange(time_dim_size)[None, :], seq_len[:, None])
 
     espnet_mhsa_module = RelPositionMultiHeadedAttention(
         num_heads=num_heads, embed_size=embed_size, dropout_rate=dropout_rate
@@ -186,7 +188,6 @@ def test_ConformerMHSARelPosV1_against_Espnet():
     )
 
     input_tensor = torch.rand((batch_dim_size, time_dim_size, embed_size))
-    sequence_mask = torch.ones((batch_dim_size, time_dim_size))
     inv_sequence_mask = torch.logical_not(sequence_mask)
 
     input_tensor_layernorm = own_mhsa_module.layernorm(input_tensor)
@@ -202,4 +203,4 @@ def test_ConformerMHSARelPosV1_against_Espnet():
 
     own_output_tensor = own_mhsa_module(input_tensor, sequence_mask=sequence_mask)
 
-    assert torch.allclose(espnet_output_tensor, own_output_tensor, rtol=1e-03)
+    assert torch.allclose(espnet_output_tensor, own_output_tensor, rtol=1e-03, atol=1e-6)