[qwen-vl] fix beam search with videos (#39726)

* fix

* fix copies

Author: zucchini-nlp

src/transformers/models/glm4v/modeling_glm4v.py

@@ -1614,14 +1614,9 @@ def _repeat_interleave_samples(x, lengths, repeat_times):
                         dict_to_expand[key], lengths=lengths, repeat_times=expand_size
                     )
                 elif key == "second_per_grid_ts":
-                    if not isinstance(dict_to_expand[key], list):
-                        raise TypeError(
-                            f"Expected value for key '{key}' to be a list, but got {type(dict_to_expand[key])} instead."
-                        )
-                    tensor = torch.tensor(dict_to_expand[key])
-                    lengths = list(video_nums)
-                    tensor = _repeat_interleave_samples(tensor, lengths=lengths, repeat_times=expand_size)
-                    dict_to_expand[key] = tensor.tolist()
+                    dict_to_expand[key] = _repeat_interleave_samples(
+                        dict_to_expand[key], lengths=list(video_nums), repeat_times=expand_size
+                    )
             return dict_to_expand
 
         def _expand_dict_for_generation(dict_to_expand):

src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py

@@ -1713,14 +1713,9 @@ def _repeat_interleave_samples(x, lengths, repeat_times):
                         dict_to_expand[key], lengths=lengths, repeat_times=expand_size
                     )
                 elif key == "second_per_grid_ts":
-                    if not isinstance(dict_to_expand[key], list):
-                        raise TypeError(
-                            f"Expected value for key '{key}' to be a list, but got {type(dict_to_expand[key])} instead."
-                        )
-                    tensor = torch.tensor(dict_to_expand[key])
-                    lengths = list(video_nums)
-                    tensor = _repeat_interleave_samples(tensor, lengths=lengths, repeat_times=expand_size)
-                    dict_to_expand[key] = tensor.tolist()
+                    dict_to_expand[key] = _repeat_interleave_samples(
+                        dict_to_expand[key], lengths=list(video_nums), repeat_times=expand_size
+                    )
             return dict_to_expand
 
         def _expand_dict_for_generation(dict_to_expand):

src/transformers/models/qwen2_vl/modeling_qwen2_vl.py

@@ -1599,14 +1599,9 @@ def _repeat_interleave_samples(x, lengths, repeat_times):
                         dict_to_expand[key], lengths=lengths, repeat_times=expand_size
                     )
                 elif key == "second_per_grid_ts":
-                    if not isinstance(dict_to_expand[key], list):
-                        raise TypeError(
-                            f"Expected value for key '{key}' to be a list, but got {type(dict_to_expand[key])} instead."
-                        )
-                    tensor = torch.tensor(dict_to_expand[key])
-                    lengths = list(video_nums)
-                    tensor = _repeat_interleave_samples(tensor, lengths=lengths, repeat_times=expand_size)
-                    dict_to_expand[key] = tensor.tolist()
+                    dict_to_expand[key] = _repeat_interleave_samples(
+                        dict_to_expand[key], lengths=list(video_nums), repeat_times=expand_size
+                    )
             return dict_to_expand
 
         def _expand_dict_for_generation(dict_to_expand):