Fix setting attention for multimodal models (#39984)

* fix

* use non-explicit `None`

* keep previously set attn if exists

Author: zucchini-nlp

src/transformers/configuration_utils.py

@@ -410,15 +410,17 @@ def _attn_implementation(self):
     def _attn_implementation(self, value: Optional[Union[str, dict]]):
         """We set it recursively on the sub-configs as well"""
         # Set if for current config
-        attn_implementation = value if not isinstance(value, dict) else value.get("", self._attn_implementation)
+        current_attn = getattr(self, "_attn_implementation", None)
+        attn_implementation = value if not isinstance(value, dict) else value.get("", current_attn)
         self._attn_implementation_internal = attn_implementation
 
         # Set it recursively on the subconfigs
         for subconfig_key in self.sub_configs:
             subconfig = getattr(self, subconfig_key, None)
             if subconfig is not None:
+                current_subconfig_attn = getattr(subconfig, "_attn_implementation", None)
                 sub_implementation = (
-                    value if not isinstance(value, dict) else value.get(subconfig_key, subconfig._attn_implementation)
+                    value if not isinstance(value, dict) else value.get(subconfig_key, current_subconfig_attn)
                 )
                 subconfig._attn_implementation = sub_implementation
 

tests/test_modeling_common.py

@@ -3680,6 +3680,20 @@ def test_attn_implementation_composite_models(self):
             model = model_class(config)
             self.assertTrue(model.config.get_text_config(decoder=True)._attn_implementation == "eager")
 
+            # Test that using `dict` atttention implementation works with `from_pretrained`
+            #  Set all backbones to "eager" because "eager" attention is always available
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                model.save_pretrained(tmpdirname)
+                new_model = model.from_pretrained(tmpdirname, attn_implementation=attn_implementation_per_subconfig)
+                self.assertTrue(new_model.config._attn_implementation == "eager")
+                for submodule in new_model.modules():
+                    if (
+                        submodule is not new_model
+                        and isinstance(submodule, PreTrainedModel)
+                        and submodule.config.__class__ != new_model.config.__class__
+                    ):
+                        self.assertTrue(submodule.config._attn_implementation == "eager")
+
     @require_torch_sdpa
     def test_sdpa_can_dispatch_non_composite_models(self):
         """