wehub-resource-sync
292 lines
10 KiB
Python
292 lines
10 KiB
Python
# Copyright (c) 2023 PaddlePaddle Authors. 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 random
|
|
|
|
import numpy as np
|
|
import paddle
|
|
|
|
from paddlenlp.transformers import (
|
|
GLMConfig,
|
|
GLMForConditionalGeneration,
|
|
GLMForMultipleChoice,
|
|
GLMModel,
|
|
GLMTokenizer,
|
|
)
|
|
from tests.testing_utils import PaddleNLPModelTest, slow
|
|
from tests.transformers.test_generation_utils import GenerationTesterMixin
|
|
from tests.transformers.test_modeling_common import (
|
|
ModelTesterMixin,
|
|
ids_tensor,
|
|
random_attention_mask,
|
|
)
|
|
|
|
GLM_PRETRAINED_MODEL_ARCHIVE_LIST = ["THUDM/glm-515m", "THUDM/glm-2b", "THUDM/glm-large-chinese"]
|
|
|
|
|
|
class GLMModelTester:
|
|
def __init__(
|
|
self,
|
|
parent,
|
|
batch_size=14,
|
|
seq_length=7,
|
|
is_training=True,
|
|
use_attention_mask=True,
|
|
use_position_ids=True,
|
|
num_layers=5,
|
|
vocab_size=99,
|
|
hidden_size=32,
|
|
num_attention_heads=4,
|
|
embedding_dropout_prob=0.1,
|
|
attention_dropout_prob=0.1,
|
|
output_dropout_prob=0.1,
|
|
max_sequence_length=512,
|
|
checkpoint_activations=False,
|
|
checkpoint_num_layers=1,
|
|
parallel_output=True,
|
|
relative_encoding=False,
|
|
block_position_encoding=True,
|
|
output_predict=False,
|
|
spell_length=None,
|
|
spell_func="lstm",
|
|
attention_scale=1.0,
|
|
initializer_range=0.02,
|
|
type_vocab_size=16,
|
|
type_sequence_label_size=2,
|
|
pool_token="cls",
|
|
num_labels=3,
|
|
num_choices=4,
|
|
scope=None,
|
|
):
|
|
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_position_ids = use_position_ids
|
|
self.num_layers = num_layers
|
|
self.vocab_size = vocab_size
|
|
self.hidden_size = hidden_size
|
|
self.num_attention_heads = num_attention_heads
|
|
self.embedding_dropout_prob = embedding_dropout_prob
|
|
self.attention_dropout_prob = attention_dropout_prob
|
|
self.output_dropout_prob = output_dropout_prob
|
|
self.max_sequence_length = max_sequence_length
|
|
self.checkpoint_activations = checkpoint_activations
|
|
self.checkpoint_num_layers = checkpoint_num_layers
|
|
self.parallel_output = parallel_output
|
|
self.relative_encoding = relative_encoding
|
|
self.block_position_encoding = block_position_encoding
|
|
self.output_predict = output_predict
|
|
self.spell_length = spell_length
|
|
self.spell_func = spell_func
|
|
self.attention_scale = attention_scale
|
|
self.initializer_range = initializer_range
|
|
self.type_vocab_size = type_vocab_size
|
|
self.type_sequence_label_size = type_sequence_label_size
|
|
self.pool_token = pool_token
|
|
self.num_labels = num_labels
|
|
self.num_choices = num_choices
|
|
self.scope = None
|
|
|
|
def prepare_config_and_inputs(self, model_class):
|
|
config = self.get_config()
|
|
|
|
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size, dtype="int64")
|
|
|
|
attention_mask = None
|
|
if self.use_attention_mask:
|
|
attention_mask = random_attention_mask(
|
|
[self.batch_size, 1, self.seq_length, self.seq_length], dtype="int64"
|
|
)
|
|
|
|
position_ids = None
|
|
if self.use_position_ids:
|
|
position_ids = paddle.arange(0, self.seq_length, dtype="int64").unsqueeze(0).unsqueeze(1)
|
|
position_ids = paddle.expand(position_ids, shape=[self.batch_size, 2, -1])
|
|
|
|
sequence_labels = None
|
|
choice_labels = None
|
|
if self.parent.use_labels:
|
|
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size, dtype="int64")
|
|
choice_labels = ids_tensor([self.batch_size], self.num_choices, dtype="int64")
|
|
|
|
return (
|
|
config,
|
|
input_ids,
|
|
position_ids,
|
|
attention_mask,
|
|
sequence_labels,
|
|
choice_labels,
|
|
)
|
|
|
|
def get_config(self):
|
|
return GLMConfig(
|
|
num_layers=self.num_layers,
|
|
vocab_size=self.vocab_size,
|
|
hidden_size=self.hidden_size,
|
|
num_attention_heads=self.num_attention_heads,
|
|
embedding_dropout_prob=self.embedding_dropout_prob,
|
|
attention_dropout_prob=self.attention_dropout_prob,
|
|
output_dropout_prob=self.output_dropout_prob,
|
|
max_sequence_length=self.max_sequence_length,
|
|
checkpoint_activations=self.checkpoint_activations,
|
|
checkpoint_num_layers=self.checkpoint_num_layers,
|
|
parallel_output=self.parallel_output,
|
|
relative_encoding=self.relative_encoding,
|
|
block_position_encoding=self.block_position_encoding,
|
|
output_predict=self.output_predict,
|
|
spell_length=self.spell_length,
|
|
spell_func=self.spell_func,
|
|
attention_scale=self.attention_scale,
|
|
initializer_range=self.initializer_range,
|
|
pool_token=self.pool_token,
|
|
use_scaled_init_for_output_weights=True,
|
|
layernorm_epsilon=1e-5,
|
|
)
|
|
|
|
def create_and_check_model(
|
|
self,
|
|
config,
|
|
input_ids,
|
|
position_ids,
|
|
attention_mask,
|
|
sequence_labels,
|
|
choice_labels,
|
|
):
|
|
model = GLMModel(config)
|
|
model.eval()
|
|
result = model(
|
|
input_ids=input_ids,
|
|
position_ids=position_ids,
|
|
attention_mask=attention_mask,
|
|
return_dict=True,
|
|
)
|
|
|
|
self.parent.assertEqual(result.last_hidden_state.shape, [self.batch_size, self.seq_length, self.hidden_size])
|
|
self.parent.assertEqual(len(result.past_key_values), config["num_layers"] + 1)
|
|
|
|
def create_and_check_for_multiple_choice(
|
|
self,
|
|
config,
|
|
input_ids,
|
|
position_ids,
|
|
attention_mask,
|
|
sequence_labels,
|
|
choice_labels,
|
|
):
|
|
self.parent.assertEqual(position_ids.shape, [self.batch_size, 2, self.seq_length])
|
|
config.output_predict = True
|
|
model = GLMForMultipleChoice(config=config)
|
|
model.eval()
|
|
choice_labels = ids_tensor([self.batch_size, self.num_choices], self.num_choices, dtype="int64")
|
|
choice_indices = paddle.to_tensor([[x for x in batch] for batch in choice_labels], dtype="int64")
|
|
choice_ids = paddle.to_tensor(
|
|
[[x for x in batch] for batch in ids_tensor(choice_labels.shape, vocab_size=self.vocab_size)],
|
|
dtype="int64",
|
|
)
|
|
|
|
result = model(
|
|
input_ids,
|
|
position_ids=position_ids,
|
|
attention_mask=attention_mask,
|
|
choice_ids=choice_ids,
|
|
choice_indices=choice_indices,
|
|
return_dict=True,
|
|
)
|
|
|
|
self.parent.assertEqual(result.logits.shape, [self.batch_size, self.num_choices])
|
|
|
|
def create_and_check_for_conditional_generation(
|
|
self,
|
|
config,
|
|
input_ids,
|
|
position_ids,
|
|
attention_mask,
|
|
sequence_labels,
|
|
choice_labels,
|
|
):
|
|
model = GLMForConditionalGeneration(config=config)
|
|
model.eval()
|
|
result = model(input_ids, attention_mask=attention_mask, position_ids=position_ids, return_dict=True)
|
|
self.parent.assertEqual(result.logits.shape, [self.batch_size, self.seq_length, self.vocab_size])
|
|
self.parent.assertEqual(len(result.past_key_values), self.num_layers + 1)
|
|
self.parent.assertEqual(result.past_key_values[0].shape, [self.seq_length, self.seq_length, self.hidden_size])
|
|
|
|
def prepare_config_and_inputs_for_common(self):
|
|
config_and_inputs = self.prepare_config_and_inputs("GLMModel")
|
|
(
|
|
config,
|
|
input_ids,
|
|
position_ids,
|
|
attention_mask,
|
|
sequence_labels,
|
|
choice_labels,
|
|
) = config_and_inputs
|
|
|
|
input_dict = {
|
|
"input_ids": input_ids,
|
|
"position_ids": position_ids,
|
|
"attention_mask": attention_mask,
|
|
}
|
|
|
|
return config, input_dict
|
|
|
|
|
|
class GLMModelTest(ModelTesterMixin, GenerationTesterMixin, PaddleNLPModelTest):
|
|
base_model_class = GLMModel
|
|
use_labels = False
|
|
return_dict = False
|
|
|
|
all_model_classes = (GLMModel,)
|
|
all_generative_model_classes = {}
|
|
test_missing_keys = False
|
|
test_model_parallel = True
|
|
use_test_input_embeds = False
|
|
|
|
def setUp(self):
|
|
self.model_tester = GLMModelTester(self)
|
|
random.seed(128)
|
|
np.random.seed(128)
|
|
paddle.seed(128)
|
|
|
|
def test_glm_model(self):
|
|
config_and_inputs = self.model_tester.prepare_config_and_inputs("GLMModel")
|
|
self.model_tester.create_and_check_model(*config_and_inputs)
|
|
|
|
def test_for_multiple_choice(self):
|
|
config_and_inputs = self.model_tester.prepare_config_and_inputs("GLMForMultipleChoice")
|
|
self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
|
|
|
|
def test_for_conditional_generation(self):
|
|
config_and_inputs = self.model_tester.prepare_config_and_inputs("GLMForConditionalGeneration")
|
|
self.model_tester.create_and_check_for_conditional_generation(*config_and_inputs)
|
|
|
|
@slow
|
|
def test_model_from_pretrained(self):
|
|
for model_name in GLM_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
|
|
model = GLMModel.from_pretrained(model_name)
|
|
tokenizer = GLMTokenizer.from_pretrained(model_name)
|
|
tokens = tokenizer("hello world [MASK]")
|
|
input_ids = tokens["input_ids"]
|
|
position_ids = tokens["position_ids"]
|
|
attention_mask = tokens["attention_mask"]
|
|
|
|
input_ids = paddle.to_tensor([input_ids])
|
|
position_ids = paddle.to_tensor([position_ids])
|
|
attention_mask = paddle.to_tensor([attention_mask])
|
|
model(input_ids=input_ids, position_ids=position_ids, attention_mask=attention_mask)
|
|
self.assertIsNotNone(model)
|