def test_inference_no_head(self):
model = TFRobertaModel.from_pretrained("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(
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.num_layers = config.num_labels
self.backbone = TFRobertaModel(config,
*inputs,
**kwargs,
name="roberta_backbone")
self.dropout = tf.keras.layers.Dropout(0.2)
self.dropout_multisampled = tf.keras.layers.Dropout(0.5)
self.classifiers = [
tf.keras.layers.Dense(
1,
kernel_initializer=get_initializer(config.initializer_range),
name="classifier") for _ in range(config.num_labels)
]
self.concat = tf.keras.layers.Concatenate(axis=-1)
self.hidden_states_weights = tf.Variable(
initial_value=[-3.0] * config.num_hidden_layers + [0.0],
dtype='float32',
trainable=True,
name="hidden_state_weights")
self.softmax_act = tf.keras.layers.Softmax(axis=0)
self.backbone.roberta.pooler._trainable = False
def test_model_from_pretrained(self):
cache_dir = "/tmp/transformers_test/"
for model_name in list(
TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
model = TFRobertaModel.from_pretrained(model_name,
cache_dir=cache_dir)
shutil.rmtree(cache_dir)
self.assertIsNotNone(model)
def test_inference_no_head(self):
model = TFRobertaModel.from_pretrained("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))
def create_and_check_roberta_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}
sequence_output = model(inputs)[0]
inputs = [input_ids, input_mask]
sequence_output = model(inputs)[0]
sequence_output = model(input_ids)[0]
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_roberta_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 __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.num_layers = config.num_labels
self.backbone = TFRobertaModel(config,
*inputs,
**kwargs,
name="roberta_backbone")
self.dropout = tf.keras.layers.Dropout(0.2)
self.dropout_multisampled = tf.keras.layers.Dropout(0.5)
self.weighted_sum = WeightedSumLayer(config.num_hidden_layers)
self.classifier = tf.keras.layers.Dense(
config.num_labels,
kernel_initializer=get_initializer(config.initializer_range),
name="classifier")
self.backbone.roberta.pooler._trainable = False
def test_model_from_pretrained(self):
for model_name in list(TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
model = TFRobertaModel.from_pretrained(model_name, cache_dir=CACHE_DIR)
self.assertIsNotNone(model)
def test_model_from_pretrained(self):
for model_name in TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = TFRobertaModel.from_pretrained(model_name)
self.assertIsNotNone(model)
