Generation tests: update imagegpt input name, remove unused functions (#33663)

src/transformers/models/imagegpt/modeling_imagegpt.py

@@ -690,8 +690,7 @@ class ImageGPTModel(ImageGPTPreTrainedModel):
 
         if "pixel_values" in kwargs:
             warnings.warn(
-                "The `pixel_values` argument is deprecated and will be removed in a future version, use `input_ids`"
+                "The `pixel_values` argument is deprecated and will be removed in v4.47, use `input_ids` instead.",
-                " instead.",
                 FutureWarning,
             )
 
@@ -1004,8 +1003,7 @@ class ImageGPTForCausalImageModeling(ImageGPTPreTrainedModel, GenerationMixin):
 
         if "pixel_values" in kwargs:
             warnings.warn(
-                "The `pixel_values` argument is deprecated and will be removed in a future version, use `input_ids`"
+                "The `pixel_values` argument is deprecated and will be removed in v4.47, use `input_ids` instead.",
-                " instead.",
                 FutureWarning,
             )
 
@@ -1137,8 +1135,7 @@ class ImageGPTForImageClassification(ImageGPTPreTrainedModel):
 
         if "pixel_values" in kwargs:
             warnings.warn(
-                "The `pixel_values` argument is deprecated and will be removed in a future version, use `input_ids`"
+                "The `pixel_values` argument is deprecated and will be removed in v4.47, use `input_ids` instead.",
-                " instead.",
                 FutureWarning,
             )
 

tests/generation/test_utils.py

@@ -190,26 +190,6 @@ class GenerationTesterMixin:
         }
         return beam_kwargs
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        generation_config = copy.deepcopy(model.generation_config)
-        model._prepare_special_tokens(generation_config)
-        input_ids = torch.zeros_like(input_ids[:, :1]) + generation_config.decoder_start_token_id
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def _greedy_generate(
         self,
         model,

tests/models/codegen/test_modeling_codegen.py

@@ -23,7 +23,7 @@ from transformers.testing_utils import backend_manual_seed, is_flaky, require_to
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -150,35 +150,6 @@ class CodeGenModelTester:
             rotary_dim=self.rotary_dim,
         )
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_codegen_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         model = CodeGenModel(config=config)
         model.to(torch_device)

tests/models/falcon_mamba/test_modeling_falcon_mamba.py

@@ -150,25 +150,6 @@ class FalconMambaModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            attention_mask,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        return (
-            config,
-            input_ids,
-            attention_mask,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        )
-
     def create_and_check_falcon_mamba_model(self, config, input_ids, *args):
         config.output_hidden_states = True
         model = FalconMambaModel(config=config)

tests/models/gpt_bigcode/test_modeling_gpt_bigcode.py

@@ -22,7 +22,7 @@ from transformers.testing_utils import require_torch, slow, torch_device
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -178,35 +178,6 @@ class GPTBigCodeModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_gpt_bigcode_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         model = GPTBigCodeModel(config=config)
         model.to(torch_device)

tests/models/gpt_neo/test_modeling_gpt_neo.py

@@ -22,7 +22,7 @@ from transformers.utils import cached_property
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -157,35 +157,6 @@ class GPTNeoModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_gpt_neo_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         model = GPTNeoModel(config=config)
         model.to(torch_device)

tests/models/gptj/test_modeling_gptj.py

@@ -32,7 +32,7 @@ from transformers.testing_utils import (
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -173,35 +173,6 @@ class GPTJModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_gptj_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         model = GPTJModel(config=config)
         model.to(torch_device)

tests/models/imagegpt/test_modeling_imagegpt.py

@@ -14,10 +14,7 @@
 # limitations under the License.
 
 
-import copy
 import inspect
-import os
-import tempfile
 import unittest
 
 from transformers import ImageGPTConfig
@@ -26,13 +23,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import (
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
-    ModelTesterMixin,
-    _config_zero_init,
-    floats_tensor,
-    ids_tensor,
-    random_attention_mask,
-)
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -108,7 +99,7 @@ class ImageGPTModelTester:
     def prepare_config_and_inputs(
         self, gradient_checkpointing=False, scale_attn_by_inverse_layer_idx=False, reorder_and_upcast_attn=False
     ):
-        pixel_values = ids_tensor([self.batch_size, self.seq_length], self.vocab_size - 1)
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size - 1)
 
         input_mask = None
         if self.use_input_mask:
@@ -140,7 +131,7 @@ class ImageGPTModelTester:
 
         return (
             config,
-            pixel_values,
+            input_ids,
             input_mask,
             head_mask,
             token_type_ids,
@@ -177,66 +168,37 @@ class ImageGPTModelTester:
         config.max_position_embeddings = 1024
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
+    def create_and_check_imagegpt_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
-        (
-            config,
-            pixel_values,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            pixel_values,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
-    def create_and_check_imagegpt_model(self, config, pixel_values, input_mask, head_mask, token_type_ids, *args):
         model = ImageGPTModel(config=config)
         model.to(torch_device)
         model.eval()
 
-        result = model(pixel_values, token_type_ids=token_type_ids, head_mask=head_mask)
+        result = model(input_ids, token_type_ids=token_type_ids, head_mask=head_mask)
-        result = model(pixel_values, token_type_ids=token_type_ids)
+        result = model(input_ids, token_type_ids=token_type_ids)
-        result = model(pixel_values)
+        result = model(input_ids)
 
         self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
         self.parent.assertEqual(len(result.past_key_values), config.n_layer)
 
-    def create_and_check_lm_head_model(self, config, pixel_values, input_mask, head_mask, token_type_ids, *args):
+    def create_and_check_lm_head_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         model = ImageGPTForCausalImageModeling(config)
         model.to(torch_device)
         model.eval()
 
         labels = ids_tensor([self.batch_size, self.seq_length], self.vocab_size - 1)
-        result = model(pixel_values, token_type_ids=token_type_ids, labels=labels)
+        result = model(input_ids, token_type_ids=token_type_ids, labels=labels)
         self.parent.assertEqual(result.loss.shape, ())
         # ImageGPTForCausalImageModeling doens't have tied input- and output embeddings
         self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size - 1))
 
     def create_and_check_imagegpt_for_image_classification(
-        self, config, pixel_values, input_mask, head_mask, token_type_ids, mc_token_ids, sequence_labels, *args
+        self, config, input_ids, input_mask, head_mask, token_type_ids, mc_token_ids, sequence_labels, *args
     ):
         config.num_labels = self.num_labels
         model = ImageGPTForImageClassification(config)
         model.to(torch_device)
         model.eval()
-        result = model(pixel_values, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
         self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
 
     def prepare_config_and_inputs_for_common(self):
@@ -244,7 +206,7 @@ class ImageGPTModelTester:
 
         (
             config,
-            pixel_values,
+            input_ids,
             input_mask,
             head_mask,
             token_type_ids,
@@ -255,7 +217,7 @@ class ImageGPTModelTester:
         ) = config_and_inputs
 
         inputs_dict = {
-            "pixel_values": pixel_values,
+            "input_ids": input_ids,
             "token_type_ids": token_type_ids,
             "head_mask": head_mask,
         }
@@ -275,7 +237,6 @@ class ImageGPTModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
         else {}
     )
     test_missing_keys = False
-    input_name = "pixel_values"
 
     # as ImageGPTForImageClassification isn't included in any auto mapping, we add labels here
     def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
@@ -351,228 +312,6 @@ class ImageGPTModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
             expected_arg_names = ["input_ids"]
             self.assertListEqual(arg_names[:1], expected_arg_names)
 
-    def test_resize_tokens_embeddings(self):
-        (
-            original_config,
-            inputs_dict,
-        ) = self.model_tester.prepare_config_and_inputs_for_common()
-        if not self.test_resize_embeddings:
-            self.skipTest(reason="test_resize_embeddings is set to False")
-
-        for model_class in self.all_model_classes:
-            config = copy.deepcopy(original_config)
-            model = model_class(config)
-            model.to(torch_device)
-
-            if self.model_tester.is_training is False:
-                model.eval()
-
-            model_vocab_size = config.vocab_size
-            # Retrieve the embeddings and clone theme
-            model_embed = model.resize_token_embeddings(model_vocab_size)
-            cloned_embeddings = model_embed.weight.clone()
-
-            # Check that resizing the token embeddings with a larger vocab size increases the model's vocab size
-            model_embed = model.resize_token_embeddings(model_vocab_size + 10)
-            self.assertEqual(model.config.vocab_size, model_vocab_size + 10)
-            # Check that it actually resizes the embeddings matrix
-            self.assertEqual(model_embed.weight.shape[0], cloned_embeddings.shape[0] + 10)
-            # Check that the model can still do a forward pass successfully (every parameter should be resized)
-            model(**self._prepare_for_class(inputs_dict, model_class))
-
-            # Check that resizing the token embeddings with a smaller vocab size decreases the model's vocab size
-            model_embed = model.resize_token_embeddings(model_vocab_size - 15)
-            self.assertEqual(model.config.vocab_size, model_vocab_size - 15)
-            # Check that it actually resizes the embeddings matrix
-            self.assertEqual(model_embed.weight.shape[0], cloned_embeddings.shape[0] - 15)
-
-            # Check that the model can still do a forward pass successfully (every parameter should be resized)
-            # Input ids should be clamped to the maximum size of the vocabulary
-            inputs_dict["pixel_values"].clamp_(max=model_vocab_size - 15 - 1)
-
-            # Check that adding and removing tokens has not modified the first part of the embedding matrix.
-            models_equal = True
-            for p1, p2 in zip(cloned_embeddings, model_embed.weight):
-                if p1.data.ne(p2.data).sum() > 0:
-                    models_equal = False
-
-            self.assertTrue(models_equal)
-
-    def test_resize_embeddings_untied(self):
-        (
-            original_config,
-            inputs_dict,
-        ) = self.model_tester.prepare_config_and_inputs_for_common()
-        if not self.test_resize_embeddings:
-            self.skipTest(reason="test_resize_embeddings is set to False")
-
-        original_config.tie_word_embeddings = False
-
-        # if model cannot untied embeddings -> leave test
-        if original_config.tie_word_embeddings:
-            self.skipTest(reason="tie_word_embeddings is set to False")
-
-        for model_class in self.all_model_classes:
-            config = copy.deepcopy(original_config)
-            model = model_class(config).to(torch_device)
-
-            # if no output embeddings -> leave test
-            if model.get_output_embeddings() is None:
-                continue
-
-            # Check that resizing the token embeddings with a larger vocab size increases the model's vocab size
-            model_vocab_size = config.vocab_size
-            model.resize_token_embeddings(model_vocab_size + 10)
-            self.assertEqual(model.config.vocab_size, model_vocab_size + 10)
-            output_embeds = model.get_output_embeddings()
-            self.assertEqual(output_embeds.weight.shape[0], model_vocab_size + 10)
-            # Check bias if present
-            if output_embeds.bias is not None:
-                self.assertEqual(output_embeds.bias.shape[0], model_vocab_size + 10)
-            # Check that the model can still do a forward pass successfully (every parameter should be resized)
-            model(**self._prepare_for_class(inputs_dict, model_class))
-
-            # Check that resizing the token embeddings with a smaller vocab size decreases the model's vocab size
-            model.resize_token_embeddings(model_vocab_size - 15)
-            self.assertEqual(model.config.vocab_size, model_vocab_size - 15)
-            # Check that it actually resizes the embeddings matrix
-            output_embeds = model.get_output_embeddings()
-            self.assertEqual(output_embeds.weight.shape[0], model_vocab_size - 15)
-            # Check bias if present
-            if output_embeds.bias is not None:
-                self.assertEqual(output_embeds.bias.shape[0], model_vocab_size - 15)
-            # Check that the model can still do a forward pass successfully (every parameter should be resized)
-            # Input ids should be clamped to the maximum size of the vocabulary
-            inputs_dict["pixel_values"].clamp_(max=model_vocab_size - 15 - 1)
-            # Check that the model can still do a forward pass successfully (every parameter should be resized)
-            model(**self._prepare_for_class(inputs_dict, model_class))
-
-    def test_inputs_embeds(self):
-        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-
-        for model_class in self.all_model_classes:
-            model = model_class(config)
-            model.to(torch_device)
-            model.eval()
-
-            inputs = copy.deepcopy(self._prepare_for_class(inputs_dict, model_class))
-
-            pixel_values = inputs["pixel_values"]
-            del inputs["pixel_values"]
-
-            wte = model.get_input_embeddings()
-            inputs["inputs_embeds"] = wte(pixel_values)
-
-            with torch.no_grad():
-                model(**inputs)[0]
-
-    # override because ImageGPT main input name is `pixel_values`
-    # NOTE: in latest transformers this is deprecated, `input_ids` should be used. TODO
-    def test_inputs_embeds_matches_input_ids(self):
-        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-
-        for model_class in self.all_model_classes:
-            model = model_class(config)
-            model.to(torch_device)
-            model.eval()
-
-            inputs = copy.deepcopy(self._prepare_for_class(inputs_dict, model_class))
-            with torch.no_grad():
-                out_ids = model(**inputs)[0]
-
-            pixel_values = inputs["pixel_values"]
-            del inputs["pixel_values"]
-
-            wte = model.get_input_embeddings()
-            inputs["inputs_embeds"] = wte(pixel_values)
-
-            with torch.no_grad():
-                out_embeds = model(**inputs)[0]
-
-            self.assertTrue(torch.allclose(out_embeds, out_ids))
-
-    def _create_and_check_torchscript(self, config, inputs_dict):
-        if not self.test_torchscript:
-            self.skipTest(reason="test_torchscript is set to False")
-
-        configs_no_init = _config_zero_init(config)  # To be sure we have no Nan
-        configs_no_init.torchscript = True
-        for model_class in self.all_model_classes:
-            model = model_class(config=configs_no_init)
-            model.to(torch_device)
-            model.eval()
-            inputs = self._prepare_for_class(inputs_dict, model_class)
-
-            try:
-                pixel_values = inputs["pixel_values"]
-                traced_model = torch.jit.trace(model, pixel_values)
-            except RuntimeError:
-                self.fail("Couldn't trace module.")
-
-            with tempfile.TemporaryDirectory() as tmp_dir_name:
-                pt_file_name = os.path.join(tmp_dir_name, "traced_model.pt")
-
-                try:
-                    torch.jit.save(traced_model, pt_file_name)
-                except Exception:
-                    self.fail("Couldn't save module.")
-
-                try:
-                    loaded_model = torch.jit.load(pt_file_name)
-                except Exception:
-                    self.fail("Couldn't load module.")
-
-            model.to(torch_device)
-            model.eval()
-
-            loaded_model.to(torch_device)
-            loaded_model.eval()
-
-            model_state_dict = model.state_dict()
-            loaded_model_state_dict = loaded_model.state_dict()
-
-            non_persistent_buffers = {}
-            for key in loaded_model_state_dict.keys():
-                if key not in model_state_dict.keys():
-                    non_persistent_buffers[key] = loaded_model_state_dict[key]
-
-            loaded_model_state_dict = {
-                key: value for key, value in loaded_model_state_dict.items() if key not in non_persistent_buffers
-            }
-
-            self.assertEqual(set(model_state_dict.keys()), set(loaded_model_state_dict.keys()))
-
-            model_buffers = list(model.buffers())
-            for non_persistent_buffer in non_persistent_buffers.values():
-                found_buffer = False
-                for i, model_buffer in enumerate(model_buffers):
-                    if torch.equal(non_persistent_buffer, model_buffer):
-                        found_buffer = True
-                        break
-
-                self.assertTrue(found_buffer)
-                model_buffers.pop(i)
-
-            model_buffers = list(model.buffers())
-            for non_persistent_buffer in non_persistent_buffers.values():
-                found_buffer = False
-                for i, model_buffer in enumerate(model_buffers):
-                    if torch.equal(non_persistent_buffer, model_buffer):
-                        found_buffer = True
-                        break
-
-                self.assertTrue(found_buffer)
-                model_buffers.pop(i)
-
-            models_equal = True
-            for layer_name, p1 in model_state_dict.items():
-                if layer_name in loaded_model_state_dict:
-                    p2 = loaded_model_state_dict[layer_name]
-                    if p1.data.ne(p2.data).sum() > 0:
-                        models_equal = False
-
-            self.assertTrue(models_equal)
-
     @unittest.skip(reason="The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
     def test_left_padding_compatibility(self):
         pass

tests/models/mamba/test_modeling_mamba.py

@@ -143,25 +143,6 @@ class MambaModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            attention_mask,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        return (
-            config,
-            input_ids,
-            attention_mask,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        )
-
     def create_and_check_mamba_model(self, config, input_ids, *args):
         config.output_hidden_states = True
         model = MambaModel(config=config)

tests/models/rwkv/test_modeling_rwkv.py

@@ -22,7 +22,7 @@ from transformers.testing_utils import require_torch, slow, torch_device
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -161,35 +161,6 @@ class RwkvModelTester:
         config.vocab_size = 300
         return config
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            mc_token_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            token_type_ids,
-            sequence_labels,
-            token_labels,
-            choice_labels,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_rwkv_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
         config.output_hidden_states = True
         model = RwkvModel(config=config)

tests/models/seamless_m4t/test_modeling_seamless_m4t.py

@@ -399,29 +399,6 @@ class SeamlessM4TModelWithSpeechInputTest(ModelTesterMixin, unittest.TestCase):
 
         return config, input_ids.float(), attention_mask, max_length
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        generation_config = copy.deepcopy(model.generation_config)
-        model._prepare_special_tokens(generation_config)
-        input_ids = (
-            torch.zeros(input_ids.shape[:2], dtype=torch.int64, layout=input_ids.layout, device=input_ids.device)
-            + generation_config.decoder_start_token_id
-        )
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 

tests/models/seamless_m4t_v2/test_modeling_seamless_m4t_v2.py

@@ -415,29 +415,6 @@ class SeamlessM4Tv2ModelWithSpeechInputTest(ModelTesterMixin, unittest.TestCase)
 
         return config, input_ids.float(), attention_mask, max_length
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        generation_config = copy.deepcopy(model.generation_config)
-        model._prepare_special_tokens(generation_config)
-        input_ids = (
-            torch.zeros(input_ids.shape[:2], dtype=torch.int64, layout=input_ids.layout, device=input_ids.device)
-            + generation_config.decoder_start_token_id
-        )
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 

tests/models/speech_to_text/test_modeling_speech_to_text.py

@@ -632,27 +632,6 @@ class Speech2TextModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTest
     def test_generate_without_input_ids(self):
         pass
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        input_ids = input_ids[:, :, 0]
-        generation_config = copy.deepcopy(model.generation_config)
-        model._prepare_special_tokens(generation_config)
-        input_ids = torch.zeros_like(input_ids[:, :1]) + generation_config.decoder_start_token_id
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def _check_outputs(self, output, input_ids, config, use_cache=False, num_return_sequences=1):
         batch_size, seq_length = input_ids.shape[:2]
         subsampled_seq_length = self.model_tester.get_subsampled_output_lengths(seq_length)

tests/models/speech_to_text/test_modeling_tf_speech_to_text.py

@@ -416,24 +416,6 @@ class TFSpeech2TextModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.T
     def test_generate_without_input_ids(self):
         pass
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = tf.repeat(encoder_outputs.last_hidden_state, num_interleave, axis=0)
-
-        input_ids = input_ids[:, :, 0]
-        input_ids = tf.zeros_like(input_ids[:, :1], dtype=tf.int64) + model._get_decoder_start_token_id()
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def _check_outputs(self, output, input_ids, config, use_cache=False, num_return_sequences=1):
         batch_size, seq_length = input_ids.shape[:2]
         subsampled_seq_length = self.model_tester.get_subsampled_output_lengths(seq_length)

tests/models/whisper/test_modeling_tf_whisper.py

@@ -519,26 +519,6 @@ class TFWhisperModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestC
     def test_generate_without_input_ids(self):
         pass
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        input_ids = input_ids[:, :, 0]
-        input_ids = tf.zeros_like(input_ids[:, :1], dtype=tf.int64) + tf.convert_to_tensor(
-            [model._get_decoder_start_token_id()]
-        )
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def _check_outputs(self, output, input_ids, config, use_cache=False, num_return_sequences=1):
         batch_size, mel, seq_length = input_ids.shape
         subsampled_seq_length = self.model_tester.get_subsampled_output_lengths(seq_length)

tests/models/whisper/test_modeling_whisper.py

@@ -868,26 +868,6 @@ class WhisperModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
     def test_generate_without_input_ids(self):
         pass
 
-    @staticmethod
-    def _get_encoder_outputs(
-        model, input_ids, attention_mask, output_attentions=None, output_hidden_states=None, num_interleave=1
-    ):
-        encoder = model.get_encoder()
-        encoder_outputs = encoder(
-            input_ids,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-        )
-        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.repeat_interleave(
-            num_interleave, dim=0
-        )
-        generation_config = copy.deepcopy(model.generation_config)
-        model._prepare_special_tokens(generation_config)
-        input_ids = input_ids[:, :, 0]
-        input_ids = torch.zeros_like(input_ids[:, :1], dtype=torch.long) + generation_config.decoder_start_token_id
-        attention_mask = None
-        return encoder_outputs, input_ids, attention_mask
-
     def _check_outputs(self, output, input_ids, config, use_cache=False, num_return_sequences=1):
         batch_size, mel, seq_length = input_ids.shape
         subsampled_seq_length = self.model_tester.get_subsampled_output_lengths(seq_length)
@@ -3894,13 +3874,6 @@ class WhisperStandaloneDecoderModelTester:
 
         return config, inputs_dict
 
-    def prepare_config_and_inputs_for_decoder(self):
-        config, input_features = self.prepare_config_and_inputs()
-        input_ids = input_features["input_ids"]
-        encoder_hidden_states = floats_tensor([self.batch_size, self.decoder_seq_length, self.hidden_size])
-
-        return (config, input_ids, encoder_hidden_states)
-
     def create_and_check_decoder_model_past(self, config, input_ids):
         config.use_cache = True
         model = WhisperDecoder(config=config).to(torch_device).eval()

tests/models/xglm/test_modeling_flax_xglm.py

@@ -22,7 +22,7 @@ from transformers import XGLMConfig, XGLMTokenizer, is_flax_available, is_torch_
 from transformers.testing_utils import is_pt_flax_cross_test, require_flax, require_sentencepiece, slow
 
 from ...generation.test_flax_utils import FlaxGenerationTesterMixin
-from ...test_modeling_flax_common import FlaxModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_flax_common import FlaxModelTesterMixin, ids_tensor, random_attention_mask
 
 
 if is_flax_available():
@@ -116,20 +116,6 @@ class FlaxXGLMModelTester:
         inputs_dict = {"input_ids": input_ids, "attention_mask": attention_mask}
         return config, inputs_dict
 
-    def prepare_config_and_inputs_for_decoder(self):
-        config, input_ids, attention_mask = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            attention_mask,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def check_use_cache_forward(self, model_class_name, config, input_ids, attention_mask):
         max_decoder_length = 20
         model = model_class_name(config)

tests/models/xglm/test_modeling_xglm.py

@@ -29,7 +29,7 @@ from transformers.testing_utils import (
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
-from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
 from ...test_pipeline_mixin import PipelineTesterMixin
 
 
@@ -125,26 +125,6 @@ class XGLMModelTester:
             gradient_checkpointing=gradient_checkpointing,
         )
 
-    def prepare_config_and_inputs_for_decoder(self):
-        (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-        ) = self.prepare_config_and_inputs()
-
-        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
-        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
-
-        return (
-            config,
-            input_ids,
-            input_mask,
-            head_mask,
-            encoder_hidden_states,
-            encoder_attention_mask,
-        )
-
     def create_and_check_xglm_model(self, config, input_ids, input_mask, head_mask, *args):
         model = XGLMModel(config=config)
         model.to(torch_device)