Update modeling_super_linear.py

modeling_super_linear.py

@@ -360,7 +360,7 @@ class SparseNoisyMoE(nn.Module):
         self.topk_gates = F.softmax(self.topk_values, dim=1)
         
         batch_size = x.size(0)
-        if x_enc.shape[1] < 512:
+        if x.shape[1] < 512:
             x = self.fourier_interp_dim1(x)
         expert_outputs = torch.stack([self.experts[i](x) for i in range(self.num_experts)], dim=1)