Skip to content

GitLab

Unverified Commit c67d1a02 authored by Sylvain Gugger's avatar Sylvain Gugger Committed by GitHub
Browse files

Tf model outputs (#6247) 

* TF outputs and test on BERT

* Albert to DistilBert

* All remaining TF models except T5

* Documentation

* One file forgotten

* TF outputs and test on BERT

* Albert to DistilBert

* All remaining TF models except T5

* Documentation

* One file forgotten

* Add new models and fix issues

* Quality improvements

* Add T5

* A bit of cleanup

* Fix for slow tests

* Style
parent bd0eab35
Hide whitespace changes
Inline Side-by-side
Showing with 116 additions and 275 deletions
tests/test_modeling_tf_electra.py
View file @ c67d1a02
......@@ -95,6 +95,7 @@ class TFElectraModelTester:
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
return_dict=True,
)
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -104,18 +105,15 @@ class TFElectraModelTester:
):
model = TFElectraModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(sequence_output,) = model(inputs)
result = model(inputs)
inputs = [input_ids, input_mask]
(sequence_output,) = model(inputs)
result = model(inputs)
(sequence_output,) = model(input_ids)
result = model(input_ids)
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_electra_for_masked_lm(
......@@ -123,24 +121,16 @@ class TFElectraModelTester:
):
model = TFElectraForMaskedLM(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(prediction_scores,) = model(inputs)
result = {
"prediction_scores": prediction_scores.numpy(),
}
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
def create_and_check_electra_for_pretraining(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFElectraForPreTraining(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(prediction_scores,) = model(inputs)
result = {
"prediction_scores": prediction_scores.numpy(),
}
self.parent.assertListEqual(list(result["prediction_scores"].shape), [self.batch_size, self.seq_length])
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length])
def create_and_check_electra_for_sequence_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -148,10 +138,7 @@ class TFElectraModelTester:
config.num_labels = self.num_labels
model = TFElectraForSequenceClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_labels])
def create_and_check_electra_for_multiple_choice(
......@@ -167,8 +154,7 @@ class TFElectraModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {"logits": logits.numpy()}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def create_and_check_electra_for_question_answering(
......@@ -176,11 +162,7 @@ class TFElectraModelTester:
):
model = TFElectraForQuestionAnswering(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......@@ -190,10 +172,7 @@ class TFElectraModelTester:
config.num_labels = self.num_labels
model = TFElectraForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def prepare_config_and_inputs_for_common(self):
......
tests/test_modeling_tf_flaubert.py
View file @ c67d1a02
......@@ -113,6 +113,7 @@ class TFFlaubertModelTester:
summary_type=self.summary_type,
use_proj=self.use_proj,
bos_token_id=self.bos_token_id,
return_dict=True,
)
return (
......@@ -141,16 +142,12 @@ class TFFlaubertModelTester:
):
model = TFFlaubertModel(config=config)
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
result = model(inputs)
inputs = [input_ids, input_mask]
outputs = model(inputs)
sequence_output = outputs[0]
result = {
"sequence_output": sequence_output.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_flaubert_lm_head(
......@@ -168,13 +165,7 @@ class TFFlaubertModelTester:
model = TFFlaubertWithLMHeadModel(config)
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
logits = outputs[0]
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
......@@ -194,12 +185,7 @@ class TFFlaubertModelTester:
inputs = {"input_ids": input_ids, "lengths": input_lengths}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......@@ -220,11 +206,7 @@ class TFFlaubertModelTester:
inputs = {"input_ids": input_ids, "lengths": input_lengths}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.type_sequence_label_size])
......@@ -243,10 +225,7 @@ class TFFlaubertModelTester:
config.num_labels = self.num_labels
model = TFFlaubertForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def create_and_check_flaubert_for_multiple_choice(
......@@ -271,8 +250,7 @@ class TFFlaubertModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {"logits": logits.numpy()}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def prepare_config_and_inputs_for_common(self):
......
tests/test_modeling_tf_gpt2.py
View file @ c67d1a02
......@@ -102,6 +102,7 @@ class TFGPT2ModelTester:
# initializer_range=self.initializer_range
bos_token_id=self.bos_token_id,
eos_token_id=self.eos_token_id,
return_dict=True,
)
head_mask = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2)
......@@ -125,18 +126,15 @@ class TFGPT2ModelTester:
"attention_mask": input_mask,
"token_type_ids": token_type_ids,
}
sequence_output = model(inputs)[0]
result = model(inputs)
inputs = [input_ids, None, input_mask] # None is the input for 'past'
sequence_output = model(inputs)[0]
result = model(inputs)
sequence_output = model(input_ids)[0]
result = model(input_ids)
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size],
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size],
)
def create_and_check_gpt2_model_past(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
......@@ -150,7 +148,7 @@ class TFGPT2ModelTester:
self.parent.assertTrue(len(outputs) == len(outputs_use_cache_conf))
self.parent.assertTrue(len(outputs) == len(outputs_no_past) + 1)
output, past = outputs
output, past = outputs.to_tuple()
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
......@@ -160,8 +158,8 @@ class TFGPT2ModelTester:
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
next_token_type_ids = tf.concat([token_type_ids, next_token_types], axis=-1)
output_from_no_past, _ = model(next_input_ids, token_type_ids=next_token_type_ids)
output_from_past, _ = model(next_tokens, token_type_ids=next_token_types, past=past)
output_from_no_past = model(next_input_ids, token_type_ids=next_token_type_ids)["last_hidden_state"]
output_from_past = model(next_tokens, token_type_ids=next_token_types, past=past)["last_hidden_state"]
# select random slice
random_slice_idx = int(ids_tensor((1,), shape_list(output_from_past)[-1]))
......@@ -183,7 +181,7 @@ class TFGPT2ModelTester:
attn_mask = tf.concat([attn_mask_begin, attn_mask_end], axis=1)
# first forward pass
output, past = model(input_ids, attention_mask=attn_mask)
output, past = model(input_ids, attention_mask=attn_mask).to_tuple()
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
......@@ -202,8 +200,8 @@ class TFGPT2ModelTester:
attn_mask = tf.concat([attn_mask, tf.ones((shape_list(attn_mask)[0], 1), dtype=tf.int32)], axis=1)
# get two different outputs
output_from_no_past, _ = model(next_input_ids, attention_mask=attn_mask)
output_from_past, _ = model(next_tokens, past=past, attention_mask=attn_mask)
output_from_no_past = model(next_input_ids, attention_mask=attn_mask)["last_hidden_state"]
output_from_past = model(next_tokens, past=past, attention_mask=attn_mask)["last_hidden_state"]
# select random slice
random_slice_idx = int(ids_tensor((1,), shape_list(output_from_past)[-1]))
......@@ -220,12 +218,9 @@ class TFGPT2ModelTester:
"attention_mask": input_mask,
"token_type_ids": token_type_ids,
}
prediction_scores = model(inputs)[0]
result = {
"prediction_scores": prediction_scores.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size],
list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size],
)
def create_and_check_gpt2_double_head(
......@@ -243,8 +238,7 @@ class TFGPT2ModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
lm_logits, mc_logits = model(inputs)[:2]
result = {"lm_logits": lm_logits.numpy(), "mc_logits": mc_logits.numpy()}
result = model(inputs)
self.parent.assertListEqual(
list(result["lm_logits"].shape), [self.batch_size, self.num_choices, self.seq_length, self.vocab_size],
)
......
tests/test_modeling_tf_mobilebert.py
View file @ c67d1a02
......@@ -138,6 +138,7 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
embedding_size=self.embedding_size,
return_dict=True,
)
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -147,33 +148,26 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
):
model = TFMobileBertModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
sequence_output, pooled_output = model(inputs)
result = model(inputs)
inputs = [input_ids, input_mask]
sequence_output, pooled_output = model(inputs)
result = model(inputs)
sequence_output, pooled_output = model(input_ids)
result = model(input_ids)
result = {
"sequence_output": sequence_output.numpy(),
"pooled_output": pooled_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
self.parent.assertListEqual(list(result["pooled_output"].shape), [self.batch_size, self.hidden_size])
self.parent.assertListEqual(list(result["pooler_output"].shape), [self.batch_size, self.hidden_size])
def create_and_check_mobilebert_for_masked_lm(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFMobileBertForMaskedLM(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(prediction_scores,) = model(inputs)
result = {
"prediction_scores": prediction_scores.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
def create_and_check_mobilebert_for_next_sequence_prediction(
......@@ -181,26 +175,19 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
):
model = TFMobileBertForNextSentencePrediction(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(seq_relationship_score,) = model(inputs)
result = {
"seq_relationship_score": seq_relationship_score.numpy(),
}
self.parent.assertListEqual(list(result["seq_relationship_score"].shape), [self.batch_size, 2])
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, 2])
def create_and_check_mobilebert_for_pretraining(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFMobileBertForPreTraining(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
prediction_scores, seq_relationship_score = model(inputs)
result = {
"prediction_scores": prediction_scores.numpy(),
"seq_relationship_score": seq_relationship_score.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
list(result["prediction_logits"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
self.parent.assertListEqual(list(result["seq_relationship_score"].shape), [self.batch_size, 2])
self.parent.assertListEqual(list(result["seq_relationship_logits"].shape), [self.batch_size, 2])
def create_and_check_mobilebert_for_sequence_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -208,10 +195,7 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
config.num_labels = self.num_labels
model = TFMobileBertForSequenceClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_labels])
def create_and_check_mobilebert_for_multiple_choice(
......@@ -227,10 +211,7 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def create_and_check_mobilebert_for_token_classification(
......@@ -239,10 +220,7 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
config.num_labels = self.num_labels
model = TFMobileBertForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels]
)
......@@ -252,11 +230,7 @@ class TFMobileBertModelTest(TFModelTesterMixin, unittest.TestCase):
):
model = TFMobileBertForQuestionAnswering(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......
tests/test_modeling_tf_openai_gpt.py
View file @ c67d1a02
......@@ -94,9 +94,10 @@ class TFOpenAIGPTModelTester:
# hidden_dropout_prob=self.hidden_dropout_prob,
# attention_probs_dropout_prob=self.attention_probs_dropout_prob,
n_positions=self.max_position_embeddings,
n_ctx=self.max_position_embeddings
n_ctx=self.max_position_embeddings,
# type_vocab_size=self.type_vocab_size,
# initializer_range=self.initializer_range
# initializer_range=self.initializer_range,
return_dict=True,
)
head_mask = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2)
......@@ -116,30 +117,22 @@ class TFOpenAIGPTModelTester:
def create_and_check_openai_gpt_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
model = TFOpenAIGPTModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
sequence_output = model(inputs)[0]
result = model(inputs)
inputs = [input_ids, input_mask]
sequence_output = model(inputs)[0]
result = model(inputs)
sequence_output = model(input_ids)[0]
result = model(input_ids)
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_openai_gpt_lm_head(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
model = TFOpenAIGPTLMHeadModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
prediction_scores = model(inputs)[0]
result = {
"prediction_scores": prediction_scores.numpy(),
}
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
def create_and_check_openai_gpt_double_head(
self, config, input_ids, input_mask, head_mask, token_type_ids, mc_token_ids, *args
......@@ -156,8 +149,7 @@ class TFOpenAIGPTModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
lm_logits, mc_logits = model(inputs)[:2]
result = {"lm_logits": lm_logits.numpy(), "mc_logits": mc_logits.numpy()}
result = model(inputs)
self.parent.assertListEqual(
list(result["lm_logits"].shape), [self.batch_size, self.num_choices, self.seq_length, self.vocab_size]
)
......
tests/test_modeling_tf_roberta.py
View file @ c67d1a02
......@@ -95,6 +95,7 @@ class TFRobertaModelTester:
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
return_dict=True,
)
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -104,31 +105,23 @@ class TFRobertaModelTester:
):
model = TFRobertaModel(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
sequence_output = model(inputs)[0]
result = model(inputs)
inputs = [input_ids, input_mask]
sequence_output = model(inputs)[0]
result = model(inputs)
sequence_output = model(input_ids)[0]
result = model(input_ids)
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_roberta_for_masked_lm(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = TFRobertaForMaskedLM(config=config)
prediction_scores = model([input_ids, input_mask, token_type_ids])[0]
result = {
"prediction_scores": prediction_scores.numpy(),
}
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
result = model([input_ids, input_mask, token_type_ids])
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
def create_and_check_roberta_for_token_classification(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
......@@ -136,10 +129,7 @@ class TFRobertaModelTester:
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}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def create_and_check_roberta_for_question_answering(
......@@ -147,11 +137,7 @@ class TFRobertaModelTester:
):
model = TFRobertaForQuestionAnswering(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......@@ -168,10 +154,7 @@ class TFRobertaModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def prepare_config_and_inputs_for_common(self):
......
tests/test_modeling_tf_t5.py
View file @ c67d1a02
......@@ -78,6 +78,7 @@ class TFT5ModelTester:
bos_token_id=self.pad_token_id,
pad_token_id=self.pad_token_id,
decoder_start_token_id=self.pad_token_id,
return_dict=True,
)
return (config, input_ids, input_mask, token_labels)
......@@ -89,22 +90,14 @@ class TFT5ModelTester:
"decoder_input_ids": input_ids,
"decoder_attention_mask": input_mask,
}
decoder_output, decoder_past, encoder_output = model(inputs)
decoder_output, decoder_past, encoder_output = model(
input_ids, decoder_attention_mask=input_mask, decoder_input_ids=input_ids
)
result = {
"encoder_output": encoder_output.numpy(),
"decoder_past": decoder_past,
"decoder_output": decoder_output.numpy(),
}
self.parent.assertListEqual(
list(result["encoder_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
self.parent.assertListEqual(
list(result["decoder_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
result = model(inputs)
result = model(input_ids, decoder_attention_mask=input_mask, decoder_input_ids=input_ids)
decoder_output = result["last_hidden_state"]
decoder_past = result["decoder_past_key_values"]
encoder_output = result["encoder_last_hidden_state"]
self.parent.assertListEqual(list(encoder_output.shape), [self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertListEqual(list(decoder_output.shape), [self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertEqual(len(decoder_past), 2)
# decoder_past[0] should correspond to encoder output
self.parent.assertTrue(tf.reduce_all(tf.math.equal(decoder_past[0][0], encoder_output)))
......@@ -121,14 +114,9 @@ class TFT5ModelTester:
"decoder_attention_mask": input_mask,
}
prediction_scores, _, _ = model(inputs_dict)
result = model(inputs_dict)
result = {
"prediction_scores": prediction_scores.numpy(),
}
self.parent.assertListEqual(
list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
def create_and_check_t5_decoder_model_past(self, config, input_ids, decoder_input_ids, attention_mask):
model = TFT5Model(config=config).get_decoder()
......
tests/test_modeling_tf_transfo_xl.py
View file @ c67d1a02
......@@ -79,6 +79,7 @@ class TFTransfoXLModelTester:
div_val=self.div_val,
n_layer=self.num_hidden_layers,
eos_token_id=self.eos_token_id,
return_dict=True,
)
return (config, input_ids_1, input_ids_2, lm_labels)
......@@ -90,11 +91,11 @@ class TFTransfoXLModelTester:
def create_and_check_transfo_xl_model(self, config, input_ids_1, input_ids_2, lm_labels):
model = TFTransfoXLModel(config)
hidden_states_1, mems_1 = model(input_ids_1)
hidden_states_1, mems_1 = model(input_ids_1).to_tuple()
inputs = {"input_ids": input_ids_2, "mems": mems_1}
hidden_states_2, mems_2 = model(inputs)
hidden_states_2, mems_2 = model(inputs).to_tuple()
result = {
"hidden_states_1": hidden_states_1.numpy(),
......@@ -121,16 +122,16 @@ class TFTransfoXLModelTester:
def create_and_check_transfo_xl_lm_head(self, config, input_ids_1, input_ids_2, lm_labels):
model = TFTransfoXLLMHeadModel(config)
lm_logits_1, mems_1 = model(input_ids_1)
lm_logits_1, mems_1 = model(input_ids_1).to_tuple()
inputs = {"input_ids": input_ids_1, "labels": lm_labels}
_, mems_1 = model(inputs)
_, mems_1 = model(inputs).to_tuple()
lm_logits_2, mems_2 = model([input_ids_2, mems_1])
lm_logits_2, mems_2 = model([input_ids_2, mems_1]).to_tuple()
inputs = {"input_ids": input_ids_1, "mems": mems_1, "labels": lm_labels}
_, mems_2 = model(inputs)
_, mems_2 = model(inputs).to_tuple()
result = {
"mems_1": [mem.numpy() for mem in mems_1],
......
tests/test_modeling_tf_xlm.py
View file @ c67d1a02
......@@ -112,6 +112,7 @@ class TFXLMModelTester:
summary_type=self.summary_type,
use_proj=self.use_proj,
bos_token_id=self.bos_token_id,
return_dict=True,
)
return (
......@@ -140,16 +141,12 @@ class TFXLMModelTester:
):
model = TFXLMModel(config=config)
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
result = model(inputs)
inputs = [input_ids, input_mask]
outputs = model(inputs)
sequence_output = outputs[0]
result = {
"sequence_output": sequence_output.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_xlm_lm_head(
......@@ -169,11 +166,7 @@ class TFXLMModelTester:
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
logits = outputs[0]
result = {
"logits": logits.numpy(),
}
result = outputs
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
......@@ -193,12 +186,7 @@ class TFXLMModelTester:
inputs = {"input_ids": input_ids, "lengths": input_lengths}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......@@ -219,11 +207,7 @@ class TFXLMModelTester:
inputs = {"input_ids": input_ids, "lengths": input_lengths}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.type_sequence_label_size])
......@@ -242,10 +226,7 @@ class TFXLMModelTester:
config.num_labels = self.num_labels
model = TFXLMForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def create_and_check_xlm_for_multiple_choice(
......@@ -270,8 +251,7 @@ class TFXLMModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {"logits": logits.numpy()}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def prepare_config_and_inputs_for_common(self):
......
tests/test_modeling_tf_xlm_roberta.py
View file @ c67d1a02
......@@ -36,7 +36,7 @@ class TFFlaubertModelIntegrationTest(unittest.TestCase):
"attention_mask": tf.convert_to_tensor([[1, 1, 1, 1, 1, 1]], dtype=tf.int32),
}
output = model(features)[0]
output = model(features)["last_hidden_state"]
expected_shape = tf.TensorShape((1, 6, 768))
self.assertEqual(output.shape, expected_shape)
# compare the actual values for a slice.
......
tests/test_modeling_tf_xlnet.py
View file @ c67d1a02
......@@ -110,6 +110,7 @@ class TFXLNetModelTester:
bos_token_id=self.bos_token_id,
pad_token_id=self.pad_token_id,
eos_token_id=self.eos_token_id,
return_dict=True,
)
return (
......@@ -147,17 +148,10 @@ class TFXLNetModelTester:
model = TFXLNetModel(config)
inputs = {"input_ids": input_ids_1, "input_mask": input_mask, "token_type_ids": segment_ids}
_, _ = model(inputs)
result = model(inputs)
inputs = [input_ids_1, input_mask]
outputs, mems_1 = model(inputs)
result = {
"mems_1": [mem.numpy() for mem in mems_1],
"outputs": outputs.numpy(),
}
result = model(inputs)
config.mem_len = 0
model = TFXLNetModel(config)
......@@ -165,10 +159,10 @@ class TFXLNetModelTester:
self.parent.assertEqual(len(no_mems_outputs), 1)
self.parent.assertListEqual(
list(result["outputs"].shape), [self.batch_size, self.seq_length, self.hidden_size]
list(result["last_hidden_state"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
self.parent.assertListEqual(
list(list(mem.shape) for mem in result["mems_1"]),
list(list(mem.shape) for mem in result["mems"]),
[[self.seq_length, self.batch_size, self.hidden_size]] * self.num_hidden_layers,
)
......@@ -189,16 +183,13 @@ class TFXLNetModelTester:
model = TFXLNetLMHeadModel(config)
inputs_1 = {"input_ids": input_ids_1, "token_type_ids": segment_ids}
all_logits_1, mems_1 = model(inputs_1)
all_logits_1, mems_1 = model(inputs_1).to_tuple()
inputs_2 = {"input_ids": input_ids_2, "mems": mems_1, "token_type_ids": segment_ids}
all_logits_2, mems_2 = model(inputs_2)
all_logits_2, mems_2 = model(inputs_2).to_tuple()
inputs_3 = {"input_ids": input_ids_q, "perm_mask": perm_mask, "target_mapping": target_mapping}
logits, _ = model(inputs_3)
logits, _ = model(inputs_3).to_tuple()
result = {
"mems_1": [mem.numpy() for mem in mems_1],
......@@ -240,13 +231,7 @@ class TFXLNetModelTester:
model = TFXLNetForQuestionAnsweringSimple(config)
inputs = {"input_ids": input_ids_1, "attention_mask": input_mask, "token_type_ids": segment_ids}
start_logits, end_logits, mems = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
"mems": [m.numpy() for m in mems],
}
result = model(inputs)
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
......@@ -271,16 +256,11 @@ class TFXLNetModelTester:
):
model = TFXLNetForSequenceClassification(config)
logits, mems_1 = model(input_ids_1)
result = {
"mems_1": [mem.numpy() for mem in mems_1],
"logits": logits.numpy(),
}
result = model(input_ids_1)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.type_sequence_label_size])
self.parent.assertListEqual(
list(list(mem.shape) for mem in result["mems_1"]),
list(list(mem.shape) for mem in result["mems"]),
[[self.seq_length, self.batch_size, self.hidden_size]] * self.num_hidden_layers,
)
......@@ -305,16 +285,12 @@ class TFXLNetModelTester:
"attention_mask": input_mask,
# 'token_type_ids': token_type_ids
}
logits, mems_1 = model(inputs)
result = {
"mems_1": [mem.numpy() for mem in mems_1],
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(
list(result["logits"].shape), [self.batch_size, self.seq_length, config.num_labels]
)
self.parent.assertListEqual(
list(list(mem.shape) for mem in result["mems_1"]),
list(list(mem.shape) for mem in result["mems"]),
[[self.seq_length, self.batch_size, self.hidden_size]] * self.num_hidden_layers,
)
......@@ -342,15 +318,11 @@ class TFXLNetModelTester:
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits, mems_1) = model(inputs)
result = {
"mems_1": [mem.numpy() for mem in mems_1],
"logits": logits.numpy(),
}
result = model(inputs)
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
self.parent.assertListEqual(
list(list(mem.shape) for mem in result["mems_1"]),
list(list(mem.shape) for mem in result["mems"]),
[[self.seq_length, self.batch_size * self.num_choices, self.hidden_size]] * self.num_hidden_layers,
)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment