SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[T5, TF 2.2] change tf t5 argument naming (#3547)

Browse Source 
* change tf t5 argument naming for TF 2.2

* correct bug in testing
This commit is contained in:
Patrick von Platen
2020-04-01 22:04:20 +02:00
committed by GitHub
parent 06dd597552
commit a4ee4da18a
3 changed files with 36 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
src/transformers/modeling_tf_t5.py
View File

@@ -592,8 +592,8 @@ class TFT5PreTrainedModel(TFPreTrainedModel):
input_ids = tf.constant(DUMMY_INPUTS) input_ids = tf.constant(DUMMY_INPUTS)
input_mask = tf.constant(DUMMY_MASK) input_mask = tf.constant(DUMMY_MASK)
dummy_inputs = { dummy_inputs = {
"inputs": input_ids,
"decoder_input_ids": input_ids, "decoder_input_ids": input_ids,
"input_ids": input_ids,
"decoder_attention_mask": input_mask, "decoder_attention_mask": input_mask,
} }
return dummy_inputs return dummy_inputs
@@ -637,11 +637,9 @@ T5_START_DOCSTRING = r""" The T5 model was proposed in
T5_INPUTS_DOCSTRING = r""" T5_INPUTS_DOCSTRING = r"""
Args: Args:
decoder_input_ids are usually used as a `dict` (see T5 description above for more information) containing all the following. inputs are usually used as a `dict` (see T5 description above for more information) containing all the following.
decoder_input_ids (:obj:`tf.Tensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`, defaults to :obj:`None`):
Provide for sequence to sequence training. T5 uses the pad_token_id as the starting token for decoder_input_ids generation.
input_ids (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`): inputs (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`):
Indices of input sequence tokens in the vocabulary. Indices of input sequence tokens in the vocabulary.
T5 is a model with relative position embeddings so you should be able to pad the inputs on T5 is a model with relative position embeddings so you should be able to pad the inputs on
the right or the left. the right or the left.
@@ -650,6 +648,8 @@ T5_INPUTS_DOCSTRING = r"""
`T5 Training <./t5.html#training>`_ . `T5 Training <./t5.html#training>`_ .
See :func:`transformers.PreTrainedTokenizer.encode` and See :func:`transformers.PreTrainedTokenizer.encode` and
:func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details. :func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
decoder_input_ids (:obj:`tf.Tensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`, defaults to :obj:`None`):
Provide for sequence to sequence training. T5 uses the pad_token_id as the starting token for decoder_input_ids generation.
attention_mask (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`): attention_mask (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Mask to avoid performing attention on padding token indices. Mask to avoid performing attention on padding token indices.
Mask values selected in ``[0, 1]``: Mask values selected in ``[0, 1]``:
@@ -706,7 +706,7 @@ class TFT5Model(TFT5PreTrainedModel):
return self.shared return self.shared
@add_start_docstrings_to_callable(T5_INPUTS_DOCSTRING) @add_start_docstrings_to_callable(T5_INPUTS_DOCSTRING)
def call(self, decoder_input_ids, **kwargs): def call(self, inputs, **kwargs):
r""" r"""
Return: Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.T5Config`) and inputs. :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.T5Config`) and inputs.
@@ -736,13 +736,13 @@ class TFT5Model(TFT5PreTrainedModel):
""" """
if isinstance(decoder_input_ids, dict): if isinstance(inputs, dict):
kwargs.update(decoder_input_ids) kwargs.update(inputs)
else: else:
kwargs["decoder_input_ids"] = decoder_input_ids kwargs["inputs"] = inputs
# retrieve arguments # retrieve arguments
input_ids = kwargs.get("input_ids", None) input_ids = kwargs.get("inputs", None)
decoder_input_ids = kwargs.get("decoder_input_ids", None) decoder_input_ids = kwargs.get("decoder_input_ids", None)
attention_mask = kwargs.get("attention_mask", None) attention_mask = kwargs.get("attention_mask", None)
encoder_outputs = kwargs.get("encoder_outputs", None) encoder_outputs = kwargs.get("encoder_outputs", None)
@@ -803,7 +803,7 @@ class TFT5ForConditionalGeneration(TFT5PreTrainedModel):
return self.encoder return self.encoder
@add_start_docstrings_to_callable(T5_INPUTS_DOCSTRING) @add_start_docstrings_to_callable(T5_INPUTS_DOCSTRING)
def call(self, decoder_input_ids, **kwargs): def call(self, inputs, **kwargs):
r""" r"""
Return: Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.T5Config`) and inputs. :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.T5Config`) and inputs.
@@ -839,13 +839,13 @@ class TFT5ForConditionalGeneration(TFT5PreTrainedModel):
""" """
if isinstance(decoder_input_ids, dict): if isinstance(inputs, dict):
kwargs.update(decoder_input_ids) kwargs.update(inputs)
else: else:
kwargs["decoder_input_ids"] = decoder_input_ids kwargs["inputs"] = inputs
# retrieve arguments # retrieve arguments
input_ids = kwargs.get("input_ids", None) input_ids = kwargs.get("inputs", None)
decoder_input_ids = kwargs.get("decoder_input_ids", None) decoder_input_ids = kwargs.get("decoder_input_ids", None)
attention_mask = kwargs.get("attention_mask", None) attention_mask = kwargs.get("attention_mask", None)
encoder_outputs = kwargs.get("encoder_outputs", None) encoder_outputs = kwargs.get("encoder_outputs", None)
@@ -890,7 +890,8 @@ class TFT5ForConditionalGeneration(TFT5PreTrainedModel):
encoder_outputs = (past,) encoder_outputs = (past,)
return { return {
"inputs": input_ids, "inputs": None, # inputs don't have to be defined, but still need to be passed to make Keras.layer.__call__ happy
"decoder_input_ids": input_ids, # input_ids are the decoder_input_ids
"encoder_outputs": encoder_outputs, "encoder_outputs": encoder_outputs,
"attention_mask": attention_mask, "attention_mask": attention_mask,
} }

18
tests/test_modeling_tf_common.py
View File

@@ -162,6 +162,10 @@ class TFModelTesterMixin:
pt_inputs_dict = dict( pt_inputs_dict = dict(
(name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items() (name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items()
) )
# need to rename encoder-decoder "inputs" for PyTorch
if "inputs" in pt_inputs_dict and self.is_encoder_decoder:
pt_inputs_dict["input_ids"] = pt_inputs_dict.pop("inputs")
with torch.no_grad(): with torch.no_grad():
pto = pt_model(**pt_inputs_dict) pto = pt_model(**pt_inputs_dict)
tfo = tf_model(inputs_dict, training=False) tfo = tf_model(inputs_dict, training=False)
@@ -201,6 +205,10 @@ class TFModelTesterMixin:
pt_inputs_dict = dict( pt_inputs_dict = dict(
(name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items() (name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items()
) )
# need to rename encoder-decoder "inputs" for PyTorch
if "inputs" in pt_inputs_dict and self.is_encoder_decoder:
pt_inputs_dict["input_ids"] = pt_inputs_dict.pop("inputs")
with torch.no_grad(): with torch.no_grad():
pto = pt_model(**pt_inputs_dict) pto = pt_model(**pt_inputs_dict)
tfo = tf_model(inputs_dict) tfo = tf_model(inputs_dict)
@@ -223,7 +231,7 @@ class TFModelTesterMixin:
if self.is_encoder_decoder: if self.is_encoder_decoder:
input_ids = { input_ids = {
"decoder_input_ids": tf.keras.Input(batch_shape=(2, 2000), name="decoder_input_ids", dtype="int32"), "decoder_input_ids": tf.keras.Input(batch_shape=(2, 2000), name="decoder_input_ids", dtype="int32"),
"input_ids": tf.keras.Input(batch_shape=(2, 2000), name="input_ids", dtype="int32"), "inputs": tf.keras.Input(batch_shape=(2, 2000), name="inputs", dtype="int32"),
} }
else: else:
input_ids = tf.keras.Input(batch_shape=(2, 2000), name="input_ids", dtype="int32") input_ids = tf.keras.Input(batch_shape=(2, 2000), name="input_ids", dtype="int32")
@@ -259,7 +267,7 @@ class TFModelTesterMixin:
outputs_dict = model(inputs_dict) outputs_dict = model(inputs_dict)
inputs_keywords = copy.deepcopy(inputs_dict) inputs_keywords = copy.deepcopy(inputs_dict)
input_ids = inputs_keywords.pop("input_ids" if not self.is_encoder_decoder else "decoder_input_ids", None,) input_ids = inputs_keywords.pop("input_ids" if not self.is_encoder_decoder else "inputs", None,)
outputs_keywords = model(input_ids, **inputs_keywords) outputs_keywords = model(input_ids, **inputs_keywords)
output_dict = outputs_dict[0].numpy() output_dict = outputs_dict[0].numpy()
@@ -395,9 +403,9 @@ class TFModelTesterMixin:
input_ids = inputs_dict["input_ids"] input_ids = inputs_dict["input_ids"]
del inputs_dict["input_ids"] del inputs_dict["input_ids"]
else: else:
encoder_input_ids = inputs_dict["input_ids"] encoder_input_ids = inputs_dict["inputs"]
decoder_input_ids = inputs_dict["decoder_input_ids"] decoder_input_ids = inputs_dict["decoder_input_ids"]
del inputs_dict["input_ids"] del inputs_dict["inputs"]
del inputs_dict["decoder_input_ids"] del inputs_dict["decoder_input_ids"]
for model_class in self.all_model_classes: for model_class in self.all_model_classes:
@@ -415,7 +423,7 @@ class TFModelTesterMixin:
def test_lm_head_model_random_generate(self): def test_lm_head_model_random_generate(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common() config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict["input_ids"] input_ids = inputs_dict["input_ids"] if "input_ids" in inputs_dict else inputs_dict["inputs"]
if self.is_encoder_decoder: if self.is_encoder_decoder:
config.output_past = True # needed for Bart TODO: might have to update for other encoder-decoder models config.output_past = True # needed for Bart TODO: might have to update for other encoder-decoder models

10
tests/test_modeling_tf_t5.py
View File

@@ -107,13 +107,15 @@ class TFT5ModelTest(TFModelTesterMixin, unittest.TestCase):
def create_and_check_t5_model(self, config, input_ids, input_mask, token_labels): def create_and_check_t5_model(self, config, input_ids, input_mask, token_labels):
model = TFT5Model(config=config) model = TFT5Model(config=config)
inputs = { inputs = {
"input_ids": input_ids, "inputs": input_ids,
"decoder_input_ids": input_ids, "decoder_input_ids": input_ids,
"decoder_attention_mask": input_mask, "decoder_attention_mask": input_mask,
} }
encoder_output, decoder_output = model(inputs) encoder_output, decoder_output = model(inputs)
encoder_output, decoder_output = model(input_ids, decoder_attention_mask=input_mask, input_ids=input_ids) encoder_output, decoder_output = model(
input_ids, decoder_attention_mask=input_mask, decoder_input_ids=input_ids
)
result = { result = {
"encoder_output": encoder_output.numpy(), "encoder_output": encoder_output.numpy(),
@@ -129,7 +131,7 @@ class TFT5ModelTest(TFModelTesterMixin, unittest.TestCase):
def create_and_check_t5_with_lm_head(self, config, input_ids, input_mask, token_labels): def create_and_check_t5_with_lm_head(self, config, input_ids, input_mask, token_labels):
model = TFT5ForConditionalGeneration(config=config) model = TFT5ForConditionalGeneration(config=config)
inputs_dict = { inputs_dict = {
"input_ids": input_ids, "inputs": input_ids,
"decoder_input_ids": input_ids, "decoder_input_ids": input_ids,
"decoder_attention_mask": input_mask, "decoder_attention_mask": input_mask,
} }
@@ -147,7 +149,7 @@ class TFT5ModelTest(TFModelTesterMixin, unittest.TestCase):
config_and_inputs = self.prepare_config_and_inputs() config_and_inputs = self.prepare_config_and_inputs()
(config, input_ids, input_mask, token_labels) = config_and_inputs (config, input_ids, input_mask, token_labels) = config_and_inputs
inputs_dict = { inputs_dict = {
"input_ids": input_ids, "inputs": input_ids,
"decoder_input_ids": input_ids, "decoder_input_ids": input_ids,
"decoder_attention_mask": input_mask, "decoder_attention_mask": input_mask,
} }