Add T5 Encoder for Feature Extraction (#8717)

* Add T5 Encoder class for feature extraction * fix T5 encoder add_start_docstrings indent * update init with T5 encoder * update init with TFT5ModelEncoder * remove TFT5ModelEncoder * change T5ModelEncoder order in init * add T5ModelEncoder to transformers init * clean T5ModelEncoder * update init with TFT5ModelEncoder * add TFModelEncoder for Tensorflow * update init with TFT5ModelEncoder * Update src/transformers/models/t5/modeling_t5.py change output from Seq2SeqModelOutput to BaseModelOutput Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> * remove encoder_outputs 1. remove encoder_outputs from the function call. 2. remove the encoder_outputs If statement. 3. remove isinstance from return_dict. * Authorize missing decoder keys * remove unnecessary input parameters remove pask_key_values and use_cache * remove use_cache remove use_cache from the forward method * add doctoring for T5 encoder add doctoring for T5 encoder with T5_ENCODER_INPUTS_DOCSTRING * change return_dict to dot access * add T5_ENCODER_INPUTS_DOCSTRING for TF T5 * change TFT5Encoder output type to BaseModelOutput * remove unnecessary parameters for TFT5Encoder * remove unnecessary if statement * add import BaseModelOutput * fix BaseModelOutput typo to TFBaseModelOutput * update T5 doc with T5ModelEncoder * add T5ModelEncoder to tests * finish pytorch * finish docs and mt5 * add mtf to init * fix init * remove n_positions * finish PR * Update src/transformers/models/mt5/modeling_mt5.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> * Update src/transformers/models/t5/modeling_t5.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> * Update src/transformers/models/t5/modeling_tf_t5.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> * Update src/transformers/models/mt5/modeling_tf_mt5.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> * make style Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
2020-11-30 08:34:40 +01:00
parent 610cb106a2
commit 40ecaf0c2b
15 changed files with 656 additions and 18 deletions
--- a/tests/test_modeling_rag.py
+++ b/tests/test_modeling_rag.py
@@ -554,7 +554,7 @@ class RagDPRT5Test(RagTestMixin, unittest.TestCase):
    def config_and_inputs(self):
        question_encoder_tester = DPRModelTester(self)
        dpr_config_and_inputs = question_encoder_tester.prepare_config_and_inputs()
-        generator_tester = T5ModelTester(self, vocab_size=1100, n_positions=30)
+        generator_tester = T5ModelTester(self, vocab_size=1100)
        t5_config_and_inputs = generator_tester.prepare_config_and_inputs()

        (question_encoder_config, input_ids, _, input_mask, _, _, _) = dpr_config_and_inputs
--- a/tests/test_modeling_t5.py
+++ b/tests/test_modeling_t5.py
@@ -30,7 +30,7 @@ from .test_modeling_common import ModelTesterMixin, ids_tensor
 if is_torch_available():
    import torch

-    from transformers import T5Config, T5ForConditionalGeneration, T5Model, T5Tokenizer
+    from transformers import T5Config, T5EncoderModel, T5ForConditionalGeneration, T5Model, T5Tokenizer
    from transformers.models.t5.modeling_t5 import T5_PRETRAINED_MODEL_ARCHIVE_LIST


@@ -39,7 +39,6 @@ class T5ModelTester:
        self,
        parent,
        vocab_size=99,
-        n_positions=14,
        batch_size=13,
        encoder_seq_length=7,
        decoder_seq_length=9,
@@ -71,7 +70,6 @@ class T5ModelTester:
        self.use_attention_mask = use_attention_mask
        self.use_labels = use_labels
        self.vocab_size = vocab_size
-        self.n_positions = n_positions
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
@@ -104,7 +102,6 @@ class T5ModelTester:

        config = T5Config(
            vocab_size=self.vocab_size,
-            n_positions=self.n_positions,
            d_model=self.hidden_size,
            d_ff=self.d_ff,
            d_kv=self.hidden_size // self.num_attention_heads,
@@ -559,6 +556,144 @@ class T5ModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
            )


+class T5EncoderOnlyModelTester:
+    def __init__(
+        self,
+        parent,
+        vocab_size=99,
+        batch_size=13,
+        encoder_seq_length=7,
+        # For common tests
+        use_attention_mask=True,
+        hidden_size=32,
+        num_hidden_layers=5,
+        num_attention_heads=4,
+        d_ff=37,
+        relative_attention_num_buckets=8,
+        is_training=False,
+        dropout_rate=0.1,
+        initializer_factor=0.002,
+        is_encoder_decoder=False,
+        eos_token_id=1,
+        pad_token_id=0,
+        scope=None,
+    ):
+
+        self.parent = parent
+        self.batch_size = batch_size
+        self.encoder_seq_length = encoder_seq_length
+        # For common tests
+        self.seq_length = self.encoder_seq_length
+        self.use_attention_mask = use_attention_mask
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.d_ff = d_ff
+        self.relative_attention_num_buckets = relative_attention_num_buckets
+        self.dropout_rate = dropout_rate
+        self.initializer_factor = initializer_factor
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.is_encoder_decoder = is_encoder_decoder
+        self.scope = None
+        self.is_training = is_training
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.encoder_seq_length], self.vocab_size)
+
+        attention_mask = None
+        if self.use_attention_mask:
+            attention_mask = ids_tensor([self.batch_size, self.encoder_seq_length], vocab_size=2)
+
+        config = T5Config(
+            vocab_size=self.vocab_size,
+            d_model=self.hidden_size,
+            d_ff=self.d_ff,
+            d_kv=self.hidden_size // self.num_attention_heads,
+            num_layers=self.num_hidden_layers,
+            num_heads=self.num_attention_heads,
+            relative_attention_num_buckets=self.relative_attention_num_buckets,
+            dropout_rate=self.dropout_rate,
+            initializer_factor=self.initializer_factor,
+            eos_token_id=self.eos_token_id,
+            bos_token_id=self.pad_token_id,
+            pad_token_id=self.pad_token_id,
+            is_encoder_decoder=self.is_encoder_decoder,
+        )
+
+        return (
+            config,
+            input_ids,
+            attention_mask,
+        )
+
+    def create_and_check_model(
+        self,
+        config,
+        input_ids,
+        attention_mask,
+    ):
+        model = T5EncoderModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+        )
+        result = model(input_ids=input_ids)
+        encoder_output = result.last_hidden_state
+
+        self.parent.assertEqual(encoder_output.size(), (self.batch_size, self.encoder_seq_length, self.hidden_size))
+
+    def create_and_check_model_fp16_forward(
+        self,
+        config,
+        input_ids,
+        attention_mask,
+    ):
+        model = T5EncoderModel(config=config).to(torch_device).half().eval()
+        output = model(input_ids, attention_mask=attention_mask)["last_hidden_state"]
+        self.parent.assertFalse(torch.isnan(output).any().item())
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            attention_mask,
+        ) = config_and_inputs
+
+        inputs_dict = {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+        }
+        return config, inputs_dict
+
+
+class T5EncoderOnlyModelTest(ModelTesterMixin, unittest.TestCase):
+    all_model_classes = (T5EncoderModel,) if is_torch_available() else ()
+    test_pruning = False
+    test_torchscript = True
+    test_resize_embeddings = False
+
+    def setUp(self):
+        self.model_tester = T5EncoderOnlyModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=T5Config, d_model=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    @unittest.skipIf(torch_device == "cpu", "Cant do half precision")
+    def test_model_fp16_forward(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model_fp16_forward(*config_and_inputs)
+
+
 def use_task_specific_params(model, task):
    model.config.update(model.config.task_specific_params[task])

--- a/tests/test_modeling_tf_t5.py
+++ b/tests/test_modeling_tf_t5.py
@@ -25,7 +25,7 @@ from .test_modeling_tf_common import TFModelTesterMixin, ids_tensor
 if is_tf_available():
    import tensorflow as tf

-    from transformers import T5Tokenizer, TFT5ForConditionalGeneration, TFT5Model
+    from transformers import T5Tokenizer, TFT5EncoderModel, TFT5ForConditionalGeneration, TFT5Model


 class TFT5ModelTester:
@@ -295,6 +295,128 @@ class TFT5ModelTest(TFModelTesterMixin, unittest.TestCase):
        pass


+class TFT5EncoderOnlyModelTester:
+    def __init__(
+        self,
+        parent,
+        vocab_size=99,
+        batch_size=13,
+        encoder_seq_length=7,
+        # For common tests
+        use_attention_mask=True,
+        hidden_size=32,
+        num_hidden_layers=5,
+        num_attention_heads=4,
+        d_ff=37,
+        relative_attention_num_buckets=8,
+        is_training=False,
+        dropout_rate=0.1,
+        initializer_factor=0.002,
+        is_encoder_decoder=False,
+        eos_token_id=1,
+        pad_token_id=0,
+        scope=None,
+    ):
+
+        self.parent = parent
+        self.batch_size = batch_size
+        self.encoder_seq_length = encoder_seq_length
+        # For common tests
+        self.seq_length = self.encoder_seq_length
+        self.use_attention_mask = use_attention_mask
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.d_ff = d_ff
+        self.relative_attention_num_buckets = relative_attention_num_buckets
+        self.dropout_rate = dropout_rate
+        self.initializer_factor = initializer_factor
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.is_encoder_decoder = is_encoder_decoder
+        self.scope = None
+        self.is_training = is_training
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.encoder_seq_length], self.vocab_size)
+
+        attention_mask = None
+        if self.use_attention_mask:
+            attention_mask = ids_tensor([self.batch_size, self.encoder_seq_length], vocab_size=2)
+
+        config = T5Config(
+            vocab_size=self.vocab_size,
+            d_model=self.hidden_size,
+            d_ff=self.d_ff,
+            d_kv=self.hidden_size // self.num_attention_heads,
+            num_layers=self.num_hidden_layers,
+            num_heads=self.num_attention_heads,
+            relative_attention_num_buckets=self.relative_attention_num_buckets,
+            dropout_rate=self.dropout_rate,
+            initializer_factor=self.initializer_factor,
+            eos_token_id=self.eos_token_id,
+            bos_token_id=self.pad_token_id,
+            pad_token_id=self.pad_token_id,
+            is_encoder_decoder=self.is_encoder_decoder,
+        )
+
+        return (
+            config,
+            input_ids,
+            attention_mask,
+        )
+
+    def create_and_check_model(
+        self,
+        config,
+        input_ids,
+        attention_mask,
+    ):
+        model = TFT5EncoderModel(config=config)
+        result = model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+        )
+        result = model(input_ids=input_ids)
+        encoder_output = result.last_hidden_state
+
+        self.parent.assertEqual(encoder_output.shape, (self.batch_size, self.encoder_seq_length, self.hidden_size))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            attention_mask,
+        ) = config_and_inputs
+
+        inputs_dict = {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+        }
+        return config, inputs_dict
+
+
+class TFT5EncoderOnlyModelTest(TFModelTesterMixin, unittest.TestCase):
+    all_model_classes = (TFT5EncoderModel,) if is_tf_available() else ()
+
+    def setUp(self):
+        self.model_tester = TFT5EncoderOnlyModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=T5Config, d_model=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    # is not able to be part of a pipeline
+    def test_train_pipeline_custom_model(self):
+        pass
+
+
@require_tf
@require_sentencepiece
@require_tokenizers