Upgrade to vllm 0.17.0 corex v4.1 overlay

vllm/model_executor/layers/logits_processor.py

@@ -61,7 +61,10 @@ class LogitsProcessor(CustomOp):
             logits = hidden_states
         else:
             # Get the logits for the next tokens.
-            logits = self._get_logits(hidden_states, lm_head, embedding_bias)
+            if hidden_states.shape[0] > 0:
+                logits = self._get_logits(hidden_states, lm_head, embedding_bias)
+            else:
+                logits = torch.empty([0, lm_head.weight.shape[0]], device=hidden_states.device, dtype=hidden_states.dtype)
         if logits is not None:
             if self.soft_cap is not None:
                 logits = logits / self.soft_cap