pt: support multitask dp test (#3573)

Fix #3471

deepmd/pt/infer/deep_eval.py

@@ -90,6 +90,7 @@ def __init__(
         *args: List[Any],
         auto_batch_size: Union[bool, int, AutoBatchSize] = True,
         neighbor_list: Optional["ase.neighborlist.NewPrimitiveNeighborList"] = None,
+        head: Optional[str] = None,
         **kwargs: Dict[str, Any],
     ):
         self.output_def = output_def
@@ -99,9 +100,24 @@ def __init__(
             if "model" in state_dict:
                 state_dict = state_dict["model"]
             self.input_param = state_dict["_extra_state"]["model_params"]
-            self.input_param["resuming"] = True
             self.multi_task = "model_dict" in self.input_param
-            assert not self.multi_task, "multitask mode currently not supported!"
+            if self.multi_task:
+                model_keys = list(self.input_param["model_dict"].keys())
+                assert (
+                    head is not None
+                ), f"Head must be set for multitask model! Available heads are: {model_keys}"
+                assert (
+                    head in model_keys
+                ), f"No head named {head} in model! Available heads are: {model_keys}"
+                self.input_param = self.input_param["model_dict"][head]
+                state_dict_head = {"_extra_state": state_dict["_extra_state"]}
+                for item in state_dict:
+                    if f"model.{head}." in item:
+                        state_dict_head[
+                            item.replace(f"model.{head}.", "model.Default.")
+                        ] = state_dict[item].clone()
+                state_dict = state_dict_head
+            self.input_param["resuming"] = True
             model = get_model(self.input_param).to(DEVICE)
             model = torch.jit.script(model)
             self.dp = ModelWrapper(model)