fix: cross encoder output (#3)

Author: rebel-sunwook

vllm_rbln/model_executor/models/optimum/encoder.py

@@ -138,6 +138,6 @@ def forward(self, model_input: ModelInputForRBLN) -> torch.Tensor:
             assert hidden_states.dim() == 2, (
                 f"We expected the shape to be dim 2 ([batch, num_labels]), "
                 f"but the current output is dim {hidden_states.dim()}.")
-            hidden_states = hidden_states[:request_nums]
+            hidden_states = hidden_states[:request_nums].squeeze(-1)
 
         return hidden_states