fix(conv): ensure tensors are contiguous before Conv1d (post-pad and post-transpose) to avoid cuDNN ptrDesc->finalize

models.py

@@ -17,6 +17,7 @@ class CausalConv1d(nn.Module):
         )
     def forward(self, x):  # x: (B, C, L)
         x = F.pad(x, (self.pad, 0))  # pad only on the left to ensure causality
+        x = x.contiguous()
         return self.conv(x)
     
 class DepthwiseSeparableCausalConvBlock(nn.Module):
@@ -29,10 +30,10 @@ class DepthwiseSeparableCausalConvBlock(nn.Module):
         self.dropout = nn.Dropout(dropout)
 
     def forward(self, x):  # x: (B, L, D)
-        y = x.transpose(1, 2)              # (B, D, L)
-        y = self.dw(y)                     # (B, D, L)
-        y = self.pw(y)                     # (B, D, L)
-        y = y.transpose(1, 2)              # (B, L, D)
+        y = x.transpose(1, 2).contiguous()  # (B, D, L)
+        y = self.dw(y)                      # (B, D, L)
+        y = self.pw(y.contiguous())         # (B, D, L)
+        y = y.transpose(1, 2).contiguous()  # (B, L, D)
         y = self.act(y)
         y = self.dropout(y)
         return self.ln(x + y)              # residual connection + layer norm (LN)