Fix: Change tensors to integers for torch.dynamo and torch.compile co…

…mpatibility (#23475)

* Fix: Change tensors to integers in torch.split() for torch.dynamo and torch.compile compatibility

* Applied the suggested fix to the utils/check_copies.py test

* Applied the suggested fix by changing the original function that gets copied

src/transformers/models/deformable_detr/modeling_deformable_detr.py

@@ -550,7 +550,7 @@ def multi_scale_deformable_attention(
 ) -> Tensor:
     batch_size, _, num_heads, hidden_dim = value.shape
     _, num_queries, num_heads, num_levels, num_points, _ = sampling_locations.shape
-    value_list = value.split([height * width for height, width in value_spatial_shapes], dim=1)
+    value_list = value.split([height.item() * width.item() for height, width in value_spatial_shapes], dim=1)
     sampling_grids = 2 * sampling_locations - 1
     sampling_value_list = []
     for level_id, (height, width) in enumerate(value_spatial_shapes):

src/transformers/models/deta/modeling_deta.py

@@ -453,7 +453,7 @@ def multi_scale_deformable_attention(
 ) -> Tensor:
     batch_size, _, num_heads, hidden_dim = value.shape
     _, num_queries, num_heads, num_levels, num_points, _ = sampling_locations.shape
-    value_list = value.split([height * width for height, width in value_spatial_shapes], dim=1)
+    value_list = value.split([height.item() * width.item() for height, width in value_spatial_shapes], dim=1)
     sampling_grids = 2 * sampling_locations - 1
     sampling_value_list = []
     for level_id, (height, width) in enumerate(value_spatial_shapes):

src/transformers/models/mask2former/modeling_mask2former.py

@@ -810,7 +810,7 @@ def multi_scale_deformable_attention(
 ) -> Tensor:
     batch_size, _, num_heads, hidden_dim = value.shape
     _, num_queries, num_heads, num_levels, num_points, _ = sampling_locations.shape
-    value_list = value.split([height * width for height, width in value_spatial_shapes], dim=1)
+    value_list = value.split([height.item() * width.item() for height, width in value_spatial_shapes], dim=1)
     sampling_grids = 2 * sampling_locations - 1
     sampling_value_list = []
     for level_id, (height, width) in enumerate(value_spatial_shapes):
@@ -1340,7 +1340,7 @@ def forward(
             else:
                 split_sizes[i] = last_hidden_state.shape[1] - level_start_index[i]
 
-        encoder_output = torch.split(last_hidden_state, split_sizes, dim=1)
+        encoder_output = torch.split(last_hidden_state, [size.item() for size in split_sizes], dim=1)
 
         # Compute final features
         outputs = [

src/transformers/models/oneformer/modeling_oneformer.py

@@ -66,7 +66,7 @@ def multi_scale_deformable_attention(
 ) -> Tensor:
     batch_size, _, num_heads, hidden_dim = value.shape
     _, num_queries, num_heads, num_levels, num_points, _ = sampling_locations.shape
-    value_list = value.split([height * width for height, width in value_spatial_shapes], dim=1)
+    value_list = value.split([height.item() * width.item() for height, width in value_spatial_shapes], dim=1)
     sampling_grids = 2 * sampling_locations - 1
     sampling_value_list = []
     for level_id, (height, width) in enumerate(value_spatial_shapes):