fix bug: dp+tp warmup

Author: Tsundoku958

lmdeploy/pytorch/engine/model_agent.py

@@ -403,6 +403,7 @@ def warmup(self):
                                             is_decoding=False,
                                             device='cuda',
                                             vocab_size=self.model_config.vocab_size)
+            inputs.build_dp_meta()
             self._forward_impl(inputs, swap_in_map=dict(), swap_out_map=dict())
 
             # warmup decoding(with cuda graph)
@@ -413,6 +414,7 @@ def warmup(self):
                                                 is_decoding=True,
                                                 device='cuda',
                                                 vocab_size=self.model_config.vocab_size)
+                inputs.build_dp_meta()
                 self._forward_impl(inputs, swap_in_map=dict(), swap_out_map=dict())
 
     async def _async_model_forward(