kbert 0.1.2

TensorFlow2.0 with Keras For BERT

BAND：BERT Application aNd Deployment

A simple and efficient BERT model training and deployment framework.

BAND：BERT Application aNd Deployment
Documents »

Examples · Report Bug · Feature Request · Questions

What is it

Encoding/Embedding is a upstream task of encoding any inputs in the form of text, image, audio, video, transactional data to fixed length vector. Embeddings are quite popular in the field of NLP, there has been various Embeddings models being proposed in recent years by researchers, some of the famous one are bert, xlnet, word2vec etc. The goal of this repo is to build one stop solution for all embeddings techniques available, here we are starting with popular text embeddings for now and later on we aim to add as much technique for image, audio, video inputs also.
Finally, embedding-as-service help you to encode any given text to fixed length vector from supported embeddings and models.

💾 Installation

Install the band via pip.

$ pip install band -U

Note that the code MUST be running on Python >= 3.6. Again module does not support Python 2!

⚡ ️Getting Started

Text Classification Example

import time
import tensorflow as tf
from transformers import BertConfig, BertTokenizer
from band.model import TFBertForSequenceClassification
from band.dataset import ChnSentiCorp
from band.progress import classification_convert_examples_to_features

USE_XLA = False
USE_AMP = False

EPOCHS = 1
BATCH_SIZE = 16
EVAL_BATCH_SIZE = 16
TEST_BATCH_SIZE = 1
MAX_SEQ_LEN = 128
LEARNING_RATE = 3e-5
SAVE_MODEL = False
pretrained_dir = "/home/band/models"

tf.config.optimizer.set_jit(USE_XLA)
tf.config.optimizer.set_experimental_options({"auto_mixed_precision": USE_AMP})

dataset = ChnSentiCorp(save_path="/tmp/band")
data, label = dataset.data, dataset.label
dataset.dataset_information()

train_number, eval_number, test_number = dataset.train_examples_num, dataset.eval_examples_num, dataset.test_examples_num

tokenizer = BertTokenizer.from_pretrained(pretrained_dir)
train_dataset = classification_convert_examples_to_features(data['train'], tokenizer, max_length=MAX_SEQ_LEN,
                                                            label_list=label,
                                                            output_mode="classification")
valid_dataset = classification_convert_examples_to_features(data['validation'], tokenizer, max_length=MAX_SEQ_LEN,
                                                            label_list=label,
                                                            output_mode="classification")
test_dataset = classification_convert_examples_to_features(data['test'], tokenizer, max_length=MAX_SEQ_LEN,
                                                           label_list=label,
                                                           output_mode="classification")

train_dataset = train_dataset.shuffle(100).batch(BATCH_SIZE, drop_remainder=True).repeat(EPOCHS)
train_dataset = train_dataset.prefetch(tf.data.experimental.AUTOTUNE)
valid_dataset = valid_dataset.batch(EVAL_BATCH_SIZE)
valid_dataset = valid_dataset.prefetch(tf.data.experimental.AUTOTUNE)
test_dataset = test_dataset.batch(TEST_BATCH_SIZE)
test_dataset = test_dataset.prefetch(tf.data.experimental.AUTOTUNE)

config = BertConfig.from_pretrained(pretrained_dir, num_labels=dataset.num_labels)
model = TFBertForSequenceClassification.from_pretrained(pretrained_dir, config=config, from_pt=True)
optimizer = tf.keras.optimizers.Adam(learning_rate=LEARNING_RATE, epsilon=1e-08)
if USE_AMP:
    optimizer = tf.keras.mixed_precision.experimental.LossScaleOptimizer(optimizer, 'dynamic')
loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
metric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')
model.compile(optimizer=optimizer, loss=loss, metrics=[metric])

history = model.fit(train_dataset, epochs=EPOCHS,
                    steps_per_epoch=train_number // BATCH_SIZE,
                    validation_data=valid_dataset,
                    validation_steps=eval_number // EVAL_BATCH_SIZE)

loss, accuracy = model.evaluate(test_dataset, steps=test_number // TEST_BATCH_SIZE)
print(loss, accuracy)

if SAVE_MODEL:
    saved_model_path = "./saved_models/{}".format(int(time.time()))
    model.save(saved_model_path, save_format="tf")

Named Entity Recognition

import time
import tensorflow as tf
from transformers import BertTokenizer, BertConfig
from band.dataset import MSRA_NER
from band.seqeval.callbacks import F1Metrics
from band.model import TFBertForTokenClassification
from band.utils import TrainConfig
from band.progress import NER_Dataset

pretrained_dir = '/home/band/models'

train_config = TrainConfig(epochs=3, train_batch_size=32, eval_batch_size=32, test_batch_size=1, max_length=128,
                           learning_rate=3e-5, save_model=False)

dataset = MSRA_NER(save_path="/tmp/band")

config = BertConfig.from_pretrained(pretrained_dir, num_labels=dataset.num_labels, return_unused_kwargs=True)
tokenizer = BertTokenizer.from_pretrained(pretrained_dir)
model = TFBertForTokenClassification.from_pretrained(pretrained_dir, config=config, from_pt=True)

ner = NER_Dataset(dataset=dataset, tokenizer=tokenizer, train_config=train_config)
model.compile(optimizer=ner.optimizer, loss=ner.loss, metrics=[ner.metric])

f1 = F1Metrics(dataset.get_labels(), validation_data=ner.valid_dataset, steps=ner.valid_steps)

history = model.fit(ner.train_dataset, epochs=train_config.epochs, steps_per_epoch=ner.test_steps, callbacks=[f1])

loss, accuracy = model.evaluate(ner.test_dataset, steps=ner.test_steps)

if train_config.save_model:
    saved_model_path = "./saved_models/{}".format(int(time.time()))
    model.save(saved_model_path, save_format="tf")

Question Answering

import time
import tensorflow as tf
from transformers import BertConfig, BertTokenizer
from band.model import TFBertForQuestionAnswering
from band.dataset import Squad
from band.progress import squad_convert_examples_to_features, parallel_squad_convert_examples_to_features

USE_XLA = False
USE_AMP = False

EPOCHS = 1
BATCH_SIZE = 4
EVAL_BATCH_SIZE = 4
TEST_BATCH_SIZE = 1
MAX_SEQ_LEN = 128
LEARNING_RATE = 3e-5
SAVE_MODEL = False
pretrained_dir = "/home/band/models"

tf.config.optimizer.set_jit(USE_XLA)
tf.config.optimizer.set_experimental_options({"auto_mixed_precision": USE_AMP})

dataset = Squad(save_path="/tmp/band")
data, label = dataset.data, dataset.label

train_number, eval_number = dataset.train_examples_num, dataset.eval_examples_num

tokenizer = BertTokenizer.from_pretrained(pretrained_dir)
train_dataset = parallel_squad_convert_examples_to_features(data['train'], tokenizer, max_length=MAX_SEQ_LEN,
                                                            doc_stride=128, is_training=True, max_query_length=64)
valid_dataset = parallel_squad_convert_examples_to_features(data['validation'], tokenizer, max_length=MAX_SEQ_LEN,
                                                            doc_stride=128, is_training=False, max_query_length=64)

train_dataset = train_dataset.shuffle(100).batch(BATCH_SIZE, drop_remainder=True).repeat(EPOCHS)
train_dataset = train_dataset.prefetch(tf.data.experimental.AUTOTUNE)
valid_dataset = valid_dataset.batch(EVAL_BATCH_SIZE)
valid_dataset = valid_dataset.prefetch(tf.data.experimental.AUTOTUNE)

config = BertConfig.from_pretrained(pretrained_dir)
model = TFBertForQuestionAnswering.from_pretrained(pretrained_dir, config=config, from_pt=True, max_length=MAX_SEQ_LEN)

print(model.summary())

optimizer = tf.keras.optimizers.Adam(learning_rate=LEARNING_RATE, epsilon=1e-08)
if USE_AMP:
    optimizer = tf.keras.mixed_precision.experimental.LossScaleOptimizer(optimizer, 'dynamic')

loss = {'start_position': tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
        'end_position': tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)}
metrics = {'start_position': tf.keras.metrics.SparseCategoricalAccuracy('accuracy'),
           'end_position': tf.keras.metrics.SparseCategoricalAccuracy('accuracy')}

model.compile(optimizer=optimizer, loss=loss, metrics=metrics)

history = model.fit(train_dataset, epochs=EPOCHS,
                    steps_per_epoch=train_number // BATCH_SIZE,
                    validation_data=valid_dataset,
                    validation_steps=eval_number // EVAL_BATCH_SIZE)

if SAVE_MODEL:
    saved_model_path = "./saved_models/{}".format(int(time.time()))
    model.save(saved_model_path, save_format="tf")

Dataset

For more information about dataset, see

Dataset Name	Language	TASK	Description
ChnSentiCorp	CN	Text Classification	Binary Classification
LCQMC	CN	Question Answer Match	Binary Classification
MSRA_NER	CN	Named Entity Recognition	Sequence Labeling
Toxic	EN	Text Classification	Multi-label Multi-label
Thucnews	CN	Text Classification	Multi-class Classification
SQUAD	EN	Machine Reading Comprehension	Span
DRCD	CN	Machine Reading Comprehension	Span
CMRC	CN	Machine Reading Comprehension	Span
GLUE	EN

✅ Supported Embeddings and Models

For more information about pretrained models, see

Stargazers over time