[trainer] fix pt loss (#7748)

* fix pt loss

* robust

* fix

* test

src/llamafactory/data/mm_plugin.py

@@ -21,7 +21,7 @@ import re
 from copy import deepcopy
 from dataclasses import dataclass
 from io import BytesIO
-from typing import TYPE_CHECKING, BinaryIO, Literal, Optional, TypedDict, Union
+from typing import TYPE_CHECKING, Any, BinaryIO, Literal, Optional, TypedDict, Union
 
 import numpy as np
 import torch
@@ -86,7 +86,7 @@ if TYPE_CHECKING:
             pass
 
 
-def _concatenate_list(input_list):
+def _concatenate_list(input_list: list[Any]) -> Union[list[Any], "NDArray", "torch.Tensor"]:
     r"""Concatenate a list of lists, numpy arrays or torch tensors.
 
     Returns: