Fix repr for conv (#32897)

add nx

src/transformers/pytorch_utils.py

@@ -96,6 +96,7 @@ class Conv1D(nn.Module):
     def __init__(self, nf, nx):
         super().__init__()
         self.nf = nf
+        self.nx = nx
         self.weight = nn.Parameter(torch.empty(nx, nf))
         self.bias = nn.Parameter(torch.zeros(nf))
         nn.init.normal_(self.weight, std=0.02)