Respect the config's attn_implementation if set (#32383)

* Respect the config's attn if set

* Update test - can override in from_config

* Fix

src/transformers/modeling_utils.py

@@ -1454,7 +1454,15 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             dtype_orig = cls._set_default_torch_dtype(torch_dtype)
 
         config = copy.deepcopy(config)  # We do not want to modify the config inplace in _from_config.
-        config._attn_implementation = kwargs.pop("attn_implementation", None)
+
+        if config._attn_implementation_internal is not None:
+            # In this case, the config has been created with the attn_implementation set by the user, which we
+            # should respect.
+            attn_implementation = config._attn_implementation_internal
+        else:
+            attn_implementation = None
+
+        config._attn_implementation = kwargs.pop("attn_implementation", attn_implementation)
         config = cls._autoset_attn_implementation(
             config,
             use_flash_attention_2=use_flash_attention_2,

tests/utils/test_modeling_utils.py

@@ -574,6 +574,60 @@ class ModelUtilsTest(TestCasePlus):
                         module.__class__.__name__, mistral_attention_classes[requested_attn_implementation]
                     )
 
+    def test_model_from_config_attn_implementation(self):
+        # test that the model can be instantiated with attn_implementation of either
+        # 1. config created with explicit attn_implementatation and from_config
+        # 2. explicit from_config's attn_implementation argument with a config argument
+        # 3. config created with explicit attn_implementatation and from_config overriding with explicit attn_implementation argument
+        attn_implementation_available = ["eager"]
+        if is_torch_sdpa_available():
+            attn_implementation_available.append("sdpa")
+
+        if is_flash_attn_2_available():
+            attn_implementation_available.append("flash_attention_2")
+
+        mistral_attention_classes = {
+            "eager": "MistralAttention",
+            "sdpa": "MistralSdpaAttention",
+            "flash_attention_2": "MistralFlashAttention2",
+        }
+        for requested_attn_implementation in attn_implementation_available:
+            config = AutoConfig.from_pretrained(TINY_MISTRAL, attn_implementation=requested_attn_implementation)
+            # Ensure the config was set correctly
+            self.assertEqual(config._attn_implementation, requested_attn_implementation)
+            self.assertEqual(config._attn_implementation_internal, requested_attn_implementation)
+            model = AutoModelForCausalLM.from_config(config)
+            self.assertEqual(model.config._attn_implementation, requested_attn_implementation)
+            for module in model.modules():
+                if "Attention" in module.__class__.__name__:
+                    self.assertEqual(
+                        module.__class__.__name__, mistral_attention_classes[requested_attn_implementation]
+                    )
+
+            config = AutoConfig.from_pretrained(TINY_MISTRAL)
+            # When the config is not set, the default is "eager"
+            self.assertEqual(config._attn_implementation, "eager")
+            self.assertEqual(config._attn_implementation_internal, None)
+            model = AutoModelForCausalLM.from_config(config=config, attn_implementation=requested_attn_implementation)
+            self.assertEqual(model.config._attn_implementation, requested_attn_implementation)
+            for module in model.modules():
+                if "Attention" in module.__class__.__name__:
+                    self.assertEqual(
+                        module.__class__.__name__, mistral_attention_classes[requested_attn_implementation]
+                    )
+
+            # Set a nonsense attn_implementation in the config, which should be overridden by the explicit argument
+            config = AutoConfig.from_pretrained(TINY_MISTRAL, attn_implementation="foo-bar-baz")
+            self.assertEqual(config._attn_implementation, "foo-bar-baz")
+            self.assertEqual(config._attn_implementation_internal, "foo-bar-baz")
+            model = AutoModelForCausalLM.from_config(config=config, attn_implementation=requested_attn_implementation)
+            self.assertEqual(model.config._attn_implementation, requested_attn_implementation)
+            for module in model.modules():
+                if "Attention" in module.__class__.__name__:
+                    self.assertEqual(
+                        module.__class__.__name__, mistral_attention_classes[requested_attn_implementation]
+                    )
+
     def test_torch_dtype_byte_sizes(self):
         torch_dtypes_and_bytes = [
             (torch.double, 8),