[modeling_utils] use less cpu memory with sharded checkpoint loading (#16844)

* less cpu memory with sharded checkpoint loading

* Trigger CI

* Trigger CI

src/transformers/modeling_utils.py

@@ -14,6 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import gc
 import json
 import os
 import re
@@ -2149,6 +2150,10 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 else:
                     error_msgs += _load_state_dict_into_model(model_to_load, state_dict, start_prefix)
 
+                # force memory release
+                del state_dict
+                gc.collect()
+
         if len(error_msgs) > 0:
             error_msg = "\n\t".join(error_msgs)
             raise RuntimeError(f"Error(s) in loading state_dict for {model.__class__.__name__}:\n\t{error_msg}")