Add logprobs to SamplingParameters in vllm (#9010)

This PR aims at amending #8806 , that I opened a few days ago, adding
the extra `logprobs` parameter that I accidentally forgot

libs/langchain/langchain/llms/vllm.py

@@ -54,6 +54,9 @@ class VLLM(BaseLLM):
     max_new_tokens: int = 512
     """Maximum number of tokens to generate per output sequence."""
 
+    logprobs: Optional[int] = None
+    """Number of log probabilities to return per output token."""
+
     client: Any  #: :meta private:
 
     @root_validator()
@@ -91,6 +94,7 @@ class VLLM(BaseLLM):
             "stop": self.stop,
             "ignore_eos": self.ignore_eos,
             "use_beam_search": self.use_beam_search,
+            "logprobs": self.logprobs,
         }
 
     def _generate(