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

Deprecate TF + JAX (#38758)

Browse Source 
* Scatter deprecation warnings around

* Delete the tests

* Make logging work properly!
This commit is contained in:
Matt
2025-06-11 17:28:06 +01:00
committed by GitHub
parent 337757cbd5
commit 9f563ada70
121 changed files with 27 additions and 50030 deletions
Download Patch File Download Diff File Expand all files Collapse all files

159
tests/models/roberta/test_modeling_flax_roberta.py
View File

@@ -1,159 +0,0 @@
# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
import numpy as np
from transformers import RobertaConfig, is_flax_available
from transformers.testing_utils import require_flax, slow
from ...test_modeling_flax_common import FlaxModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
if is_flax_available():
from transformers.models.roberta.modeling_flax_roberta import (
FlaxRobertaForCausalLM,
FlaxRobertaForMaskedLM,
FlaxRobertaForMultipleChoice,
FlaxRobertaForQuestionAnswering,
FlaxRobertaForSequenceClassification,
FlaxRobertaForTokenClassification,
FlaxRobertaModel,
)
class FlaxRobertaModelTester:
def __init__(
self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_attention_mask=True,
use_token_type_ids=True,
use_labels=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=2,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
type_sequence_label_size=2,
initializer_range=0.02,
num_choices=4,
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_attention_mask = use_attention_mask
self.use_token_type_ids = use_token_type_ids
self.use_labels = use_labels
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.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.num_choices = num_choices
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
attention_mask = None
if self.use_attention_mask:
attention_mask = random_attention_mask([self.batch_size, self.seq_length])
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
config = RobertaConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
is_decoder=False,
initializer_range=self.initializer_range,
)
return config, input_ids, token_type_ids, attention_mask
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
config, input_ids, token_type_ids, attention_mask = config_and_inputs
inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": attention_mask}
return config, inputs_dict
def prepare_config_and_inputs_for_decoder(self):
config_and_inputs = self.prepare_config_and_inputs()
config, input_ids, token_type_ids, attention_mask = config_and_inputs
config.is_decoder = True
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,
token_type_ids,
encoder_hidden_states,
encoder_attention_mask,
)
@require_flax
class FlaxRobertaModelTest(FlaxModelTesterMixin, unittest.TestCase):
test_head_masking = True
all_model_classes = (
(
FlaxRobertaModel,
FlaxRobertaForCausalLM,
FlaxRobertaForMaskedLM,
FlaxRobertaForSequenceClassification,
FlaxRobertaForTokenClassification,
FlaxRobertaForMultipleChoice,
FlaxRobertaForQuestionAnswering,
)
if is_flax_available()
else ()
)
def setUp(self):
self.model_tester = FlaxRobertaModelTester(self)
@slow
def test_model_from_pretrained(self):
for model_class_name in self.all_model_classes:
model = model_class_name.from_pretrained("FacebookAI/roberta-base", from_pt=True)
outputs = model(np.ones((1, 1)))
self.assertIsNotNone(outputs)

700
tests/models/roberta/test_modeling_tf_roberta.py
View File

@@ -1,700 +0,0 @@
# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import annotations
import unittest
from transformers import RobertaConfig, is_tf_available
from transformers.testing_utils import require_sentencepiece, require_tf, require_tokenizers, slow
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_tf_available():
import numpy
import tensorflow as tf
from transformers.models.roberta.modeling_tf_roberta import (
TFRobertaForCausalLM,
TFRobertaForMaskedLM,
TFRobertaForMultipleChoice,
TFRobertaForQuestionAnswering,
TFRobertaForSequenceClassification,
TFRobertaForTokenClassification,
TFRobertaModel,
)
class TFRobertaModelTester:
def __init__(
self,
parent,
):
self.parent = parent
self.batch_size = 13
self.seq_length = 7
self.is_training = True
self.use_input_mask = True
self.use_token_type_ids = True
self.use_labels = True
self.vocab_size = 99
self.hidden_size = 32
self.num_hidden_layers = 2
self.num_attention_heads = 4
self.intermediate_size = 37
self.hidden_act = "gelu"
self.hidden_dropout_prob = 0.1
self.attention_probs_dropout_prob = 0.1
self.max_position_embeddings = 512
self.type_vocab_size = 16
self.type_sequence_label_size = 2
self.initializer_range = 0.02
self.num_labels = 3
self.num_choices = 4
self.scope = None
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = random_attention_mask([self.batch_size, self.seq_length])
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
sequence_labels = None
token_labels = None
choice_labels = None
if self.use_labels:
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = RobertaConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
)
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
def prepare_config_and_inputs_for_decoder(self):
(
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = self.prepare_config_and_inputs()
config.is_decoder = True
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,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
)
def create_and_check_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFRobertaModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
result = model(inputs)
inputs = [input_ids, input_mask]
result = model(inputs)
result = model(input_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_causal_lm_base_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.is_decoder = True
model = TFRobertaModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
result = model(inputs)
inputs = [input_ids, input_mask]
result = model(inputs)
result = model(input_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_model_as_decoder(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
config.add_cross_attention = True
model = TFRobertaModel(config=config)
inputs = {
"input_ids": input_ids,
"attention_mask": input_mask,
"token_type_ids": token_type_ids,
"encoder_hidden_states": encoder_hidden_states,
"encoder_attention_mask": encoder_attention_mask,
}
result = model(inputs)
inputs = [input_ids, input_mask]
result = model(inputs, token_type_ids=token_type_ids, encoder_hidden_states=encoder_hidden_states)
# Also check the case where encoder outputs are not passed
result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_causal_lm_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.is_decoder = True
model = TFRobertaForCausalLM(config=config)
inputs = {
"input_ids": input_ids,
"attention_mask": input_mask,
"token_type_ids": token_type_ids,
}
prediction_scores = model(inputs)["logits"]
self.parent.assertListEqual(
list(prediction_scores.numpy().shape), [self.batch_size, self.seq_length, self.vocab_size]
)
def create_and_check_causal_lm_model_as_decoder(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
config.add_cross_attention = True
model = TFRobertaForCausalLM(config=config)
inputs = {
"input_ids": input_ids,
"attention_mask": input_mask,
"token_type_ids": token_type_ids,
"encoder_hidden_states": encoder_hidden_states,
"encoder_attention_mask": encoder_attention_mask,
}
result = model(inputs)
inputs = [input_ids, input_mask]
result = model(inputs, token_type_ids=token_type_ids, encoder_hidden_states=encoder_hidden_states)
prediction_scores = result["logits"]
self.parent.assertListEqual(
list(prediction_scores.numpy().shape), [self.batch_size, self.seq_length, self.vocab_size]
)
def create_and_check_causal_lm_model_past(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
):
config.is_decoder = True
model = TFRobertaForCausalLM(config=config)
# special to `RobertaEmbeddings` in `Roberta`:
# - its `padding_idx` and its effect on `position_ids`
# (TFRobertaEmbeddings.create_position_ids_from_input_ids)
# - `1` here is `TFRobertaEmbeddings.padding_idx`
input_ids = tf.where(input_ids == 1, 2, input_ids)
# first forward pass
outputs = model(input_ids, use_cache=True)
outputs_use_cache_conf = model(input_ids)
outputs_no_past = model(input_ids, use_cache=False)
self.parent.assertTrue(len(outputs) == len(outputs_use_cache_conf))
self.parent.assertTrue(len(outputs) == len(outputs_no_past) + 1)
past_key_values = outputs.past_key_values
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
# append to next input_ids and attn_mask
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
output_from_no_past = model(next_input_ids, output_hidden_states=True).hidden_states[0]
output_from_past = model(
next_tokens, past_key_values=past_key_values, output_hidden_states=True
).hidden_states[0]
# select random slice
random_slice_idx = int(ids_tensor((1,), output_from_past.shape[-1]))
output_from_no_past_slice = output_from_no_past[:, -1, random_slice_idx]
output_from_past_slice = output_from_past[:, 0, random_slice_idx]
# test that outputs are equal for slice
tf.debugging.assert_near(output_from_past_slice, output_from_no_past_slice, rtol=1e-6)
def create_and_check_causal_lm_model_past_with_attn_mask(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
):
config.is_decoder = True
model = TFRobertaForCausalLM(config=config)
# special to `RobertaEmbeddings` in `Roberta`:
# - its `padding_idx` and its effect on `position_ids`
# (TFRobertaEmbeddings.create_position_ids_from_input_ids)
# - `1` here is `TFRobertaEmbeddings.padding_idx`
# avoid `padding_idx` in the past
input_ids = tf.where(input_ids == 1, 2, input_ids)
# create attention mask
half_seq_length = self.seq_length // 2
attn_mask_begin = tf.ones((self.batch_size, half_seq_length), dtype=tf.int32)
attn_mask_end = tf.zeros((self.batch_size, self.seq_length - half_seq_length), dtype=tf.int32)
attn_mask = tf.concat([attn_mask_begin, attn_mask_end], axis=1)
# first forward pass
outputs = model(input_ids, attention_mask=attn_mask, use_cache=True)
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
past_key_values = outputs.past_key_values
# change a random masked slice from input_ids
random_seq_idx_to_change = ids_tensor((1,), half_seq_length).numpy() + 1
random_other_next_tokens = ids_tensor((self.batch_size, self.seq_length), config.vocab_size)
vector_condition = tf.range(self.seq_length) == (self.seq_length - random_seq_idx_to_change)
condition = tf.transpose(
tf.broadcast_to(tf.expand_dims(vector_condition, -1), (self.seq_length, self.batch_size))
)
input_ids = tf.where(condition, random_other_next_tokens, input_ids)
# avoid `padding_idx` in the past
input_ids = tf.where(input_ids == 1, 2, input_ids)
# append to next input_ids and
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
attn_mask = tf.concat(
[attn_mask, tf.ones((attn_mask.shape[0], 1), dtype=tf.int32)],
axis=1,
)
output_from_no_past = model(
next_input_ids,
attention_mask=attn_mask,
output_hidden_states=True,
).hidden_states[0]
output_from_past = model(
next_tokens, past_key_values=past_key_values, attention_mask=attn_mask, output_hidden_states=True
).hidden_states[0]
# select random slice
random_slice_idx = int(ids_tensor((1,), output_from_past.shape[-1]))
output_from_no_past_slice = output_from_no_past[:, -1, random_slice_idx]
output_from_past_slice = output_from_past[:, 0, random_slice_idx]
# test that outputs are equal for slice
tf.debugging.assert_near(output_from_past_slice, output_from_no_past_slice, rtol=1e-6)
def create_and_check_causal_lm_model_past_large_inputs(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
):
config.is_decoder = True
model = TFRobertaForCausalLM(config=config)
# special to `RobertaEmbeddings` in `Roberta`:
# - its `padding_idx` and its effect on `position_ids`
# (TFRobertaEmbeddings.create_position_ids_from_input_ids)
# - `1` here is `TFRobertaEmbeddings.padding_idx`
# avoid `padding_idx` in the past
input_ids = tf.where(input_ids == 1, 2, input_ids)
input_ids = input_ids[:1, :]
input_mask = input_mask[:1, :]
self.batch_size = 1
# first forward pass
outputs = model(input_ids, attention_mask=input_mask, use_cache=True)
past_key_values = outputs.past_key_values
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 3), config.vocab_size)
next_attn_mask = ids_tensor((self.batch_size, 3), 2)
# append to next input_ids and
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
next_attention_mask = tf.concat([input_mask, next_attn_mask], axis=-1)
output_from_no_past = model(
next_input_ids,
attention_mask=next_attention_mask,
output_hidden_states=True,
).hidden_states[0]
output_from_past = model(
next_tokens,
attention_mask=next_attention_mask,
past_key_values=past_key_values,
output_hidden_states=True,
).hidden_states[0]
self.parent.assertEqual(next_tokens.shape[1], output_from_past.shape[1])
# select random slice
random_slice_idx = int(ids_tensor((1,), output_from_past.shape[-1]))
output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx]
output_from_past_slice = output_from_past[:, :, random_slice_idx]
# test that outputs are equal for slice
tf.debugging.assert_near(output_from_past_slice, output_from_no_past_slice, rtol=1e-3)
def create_and_check_decoder_model_past_large_inputs(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
config.add_cross_attention = True
model = TFRobertaForCausalLM(config=config)
# special to `RobertaEmbeddings` in `Roberta`:
# - its `padding_idx` and its effect on `position_ids`
# (TFRobertaEmbeddings.create_position_ids_from_input_ids)
# - `1` here is `TFRobertaEmbeddings.padding_idx`
# avoid `padding_idx` in the past
input_ids = tf.where(input_ids == 1, 2, input_ids)
input_ids = input_ids[:1, :]
input_mask = input_mask[:1, :]
encoder_hidden_states = encoder_hidden_states[:1, :, :]
encoder_attention_mask = encoder_attention_mask[:1, :]
self.batch_size = 1
# first forward pass
outputs = model(
input_ids,
attention_mask=input_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
use_cache=True,
)
past_key_values = outputs.past_key_values
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 3), config.vocab_size)
next_attn_mask = ids_tensor((self.batch_size, 3), 2)
# append to next input_ids and
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
next_attention_mask = tf.concat([input_mask, next_attn_mask], axis=-1)
output_from_no_past = model(
next_input_ids,
attention_mask=next_attention_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
output_hidden_states=True,
).hidden_states[0]
output_from_past = model(
next_tokens,
attention_mask=next_attention_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
past_key_values=past_key_values,
output_hidden_states=True,
).hidden_states[0]
self.parent.assertEqual(next_tokens.shape[1], output_from_past.shape[1])
# select random slice
random_slice_idx = int(ids_tensor((1,), output_from_past.shape[-1]))
output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx]
output_from_past_slice = output_from_past[:, :, random_slice_idx]
# test that outputs are equal for slice
tf.debugging.assert_near(output_from_past_slice, output_from_no_past_slice, rtol=1e-3)
def create_and_check_for_masked_lm(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFRobertaForMaskedLM(config=config)
result = model([input_ids, input_mask, token_type_ids])
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
def create_and_check_for_token_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_labels = self.num_labels
model = TFRobertaForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
result = model(inputs)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
def create_and_check_for_question_answering(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFRobertaForQuestionAnswering(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
result = model(inputs)
self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
def create_and_check_for_multiple_choice(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_choices = self.num_choices
model = TFRobertaForMultipleChoice(config=config)
multiple_choice_inputs_ids = tf.tile(tf.expand_dims(input_ids, 1), (1, self.num_choices, 1))
multiple_choice_input_mask = tf.tile(tf.expand_dims(input_mask, 1), (1, self.num_choices, 1))
multiple_choice_token_type_ids = tf.tile(tf.expand_dims(token_type_ids, 1), (1, self.num_choices, 1))
inputs = {
"input_ids": multiple_choice_inputs_ids,
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
result = model(inputs)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_tf
class TFRobertaModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
all_model_classes = (
(
TFRobertaModel,
TFRobertaForCausalLM,
TFRobertaForMaskedLM,
TFRobertaForSequenceClassification,
TFRobertaForTokenClassification,
TFRobertaForQuestionAnswering,
)
if is_tf_available()
else ()
)
pipeline_model_mapping = (
{
"feature-extraction": TFRobertaModel,
"fill-mask": TFRobertaForMaskedLM,
"question-answering": TFRobertaForQuestionAnswering,
"text-classification": TFRobertaForSequenceClassification,
"text-generation": TFRobertaForCausalLM,
"token-classification": TFRobertaForTokenClassification,
"zero-shot": TFRobertaForSequenceClassification,
}
if is_tf_available()
else {}
)
test_head_masking = False
test_onnx = False
def setUp(self):
self.model_tester = TFRobertaModelTester(self)
self.config_tester = ConfigTester(self, config_class=RobertaConfig, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_model(self):
"""Test the base model"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_causal_lm_base_model(self):
"""Test the base model of the causal LM model
is_decoder=True, no cross_attention, no encoder outputs
"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_causal_lm_base_model(*config_and_inputs)
def test_model_as_decoder(self):
"""Test the base model as a decoder (of an encoder-decoder architecture)
is_decoder=True + cross_attention + pass encoder outputs
"""
config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_model_as_decoder(*config_and_inputs)
def test_for_masked_lm(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)
def test_for_causal_lm(self):
"""Test the causal LM model"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_causal_lm_model(*config_and_inputs)
def test_causal_lm_model_as_decoder(self):
"""Test the causal LM model as a decoder"""
config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_causal_lm_model_as_decoder(*config_and_inputs)
def test_causal_lm_model_past(self):
"""Test causal LM model with `past_key_values`"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_causal_lm_model_past(*config_and_inputs)
def test_causal_lm_model_past_with_attn_mask(self):
"""Test the causal LM model with `past_key_values` and `attention_mask`"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_causal_lm_model_past_with_attn_mask(*config_and_inputs)
def test_causal_lm_model_past_with_large_inputs(self):
"""Test the causal LM model with `past_key_values` and a longer decoder sequence length"""
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_causal_lm_model_past_large_inputs(*config_and_inputs)
def test_decoder_model_past_with_large_inputs(self):
"""Similar to `test_causal_lm_model_past_with_large_inputs` but with cross-attention"""
config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_decoder_model_past_large_inputs(*config_and_inputs)
def test_for_token_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_token_classification(*config_and_inputs)
def test_for_question_answering(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
def test_for_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
model_name = "FacebookAI/roberta-base"
model = TFRobertaModel.from_pretrained(model_name)
self.assertIsNotNone(model)
@require_tf
@require_sentencepiece
@require_tokenizers
class TFRobertaModelIntegrationTest(unittest.TestCase):
@slow
def test_inference_masked_lm(self):
model = TFRobertaForMaskedLM.from_pretrained("FacebookAI/roberta-base")
input_ids = tf.constant([[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
output = model(input_ids)[0]
expected_shape = [1, 11, 50265]
self.assertEqual(list(output.numpy().shape), expected_shape)
# compare the actual values for a slice.
expected_slice = tf.constant(
[[[33.8802, -4.3103, 22.7761], [4.6539, -2.8098, 13.6253], [1.8228, -3.6898, 8.8600]]]
)
self.assertTrue(numpy.allclose(output[:, :3, :3].numpy(), expected_slice.numpy(), atol=1e-4))
@slow
def test_inference_no_head(self):
model = TFRobertaModel.from_pretrained("FacebookAI/roberta-base")
input_ids = tf.constant([[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
output = model(input_ids)[0]
# compare the actual values for a slice.
expected_slice = tf.constant(
[[[-0.0231, 0.0782, 0.0074], [-0.1854, 0.0540, -0.0175], [0.0548, 0.0799, 0.1687]]]
)
self.assertTrue(numpy.allclose(output[:, :3, :3].numpy(), expected_slice.numpy(), atol=1e-4))
@slow
def test_inference_classification_head(self):
model = TFRobertaForSequenceClassification.from_pretrained("FacebookAI/roberta-large-mnli")
input_ids = tf.constant([[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
output = model(input_ids)[0]
expected_shape = [1, 3]
self.assertEqual(list(output.numpy().shape), expected_shape)
expected_tensor = tf.constant([[-0.9469, 0.3913, 0.5118]])
self.assertTrue(numpy.allclose(output.numpy(), expected_tensor.numpy(), atol=1e-4))