fix(vllm_client): use localhost for weight broadcasting

- Changed the host parameter from self.host to "0.0.0.0" in StatelessProcessGroup.create()
- This modification ensures that weight broadcasting works correctly in theVLLM client

Author: re-imagined

trl/extras/vllm_client.py

@@ -246,7 +246,7 @@ def init_communicator(self):
             raise Exception(f"Request failed: {response.status_code}, {response.text}")
 
         # Set up the communication group for weight broadcasting
-        pg = StatelessProcessGroup.create(host=self.host, port=self.group_port, rank=self.rank, world_size=world_size)
+        pg = StatelessProcessGroup.create(host="0.0.0.0", port=self.group_port, rank=self.rank, world_size=world_size)
         self.pynccl_comm = PyNcclCommunicator(pg, device=0)
 
     def update_named_param(self, name: str, weights: torch.Tensor):