[Bugfix] Fix compute_logits in Jamba (vllm-project#6093)

Signed-off-by: Alvant <alvasian@yandex.ru>

vllm/model_executor/models/jamba.py

@@ -876,7 +876,7 @@ def _prepare_mamba_cache(self):
 
     def compute_logits(self, hidden_states: torch.Tensor,
                        sampling_metadata: SamplingMetadata) -> torch.Tensor:
-        logits = self.logits_processor(self.lm_head.weight, hidden_states,
+        logits = self.logits_processor(self.lm_head, hidden_states,
                                        sampling_metadata)
         return logits