fastbert 0.2.2

The pipy version of FastBERT

FastBERT-pypi

The pypi version of FastBERT.

Install

Install fastbert with pip.

$ pip install fastbert

Supported Models

FastBERT-pypi is supported by the UER project, and all of UER high-quality models can be accelerated in the FastBERT way.

FastBERT object supports the following models:

Models (kernel_name)	URL	Description
google_bert_base_en	https://share.weiyun.com/fpdOtcmz	Google pretrained English BERT-base model on Wiki corpus.
google_bert_base_zh	https://share.weiyun.com/AykBph9V	Google pretrained Chinese BERT-base model on Wiki corpus.
uer_bert_large_zh	https://share.weiyun.com/chx2VhGk	UER pretrained Chinese BERT-large model on mixed corpus.
uer_bert_small_zh	https://share.weiyun.com/wZuVBM5g	UER pretrained Chinese BERT-small model on mixed corpus.
uer_bert_tiny_zh	https://share.weiyun.com/VJ3JEN9Z	UER pretrained Chinese BERT-tiny model on mixed corpus.

In fact, you don't have to download the model yourself. FastBERT will download the corresponding model file automatically at the first time you use it. If the automatically downloading failed, you can download these model files from the above URLs, and saving them to the directory of "~/.fastbert/".

Quick Start

Single sentence classification

An example of single sentence classification are shown in single_sentence_classification.

from fastbert import FastBERT

# Loading your dataset
labels = ['T', 'F']
sents_train = [
    'Do you like FastBERT?',
    'Yes, it runs faster than BERT!',
    ...
]
labels_train = [
    'T',
    'F',
    ...
]

# Creating a model
model = FastBERT(
    kernel_name="google_bert_base_en",  # "google_bert_base_zh" for Chinese
    labels=labels,
    device='cuda:0'
)

# Training the model
model.fit(
    sents_train,
    labels_train,
    model_saving_path='./fastbert.bin',
)

# Loading the model and making inference
model.load_model('./fastbert.bin')
label, exec_layers = model('I like FastBERT', speed=0.7)

Two sentences classification

An example of two sentences classification are presented in two_sentences_classification.

from fastbert import FastBERT_S2

# Loading your dataset
labels = ['T', 'F']
questions_train = [
    'FastBERT快吗?',
    '你在业务里使用FastBERT了吗?',
    ...
]
answers_train = [
    '快！而且速度还可调.',
    '用了啊，帮我省了好几百台机器.',
    ...
]
labels_train = [
    'T',
    'T',
    ...
]

# Creating a model
model = FastBERT_S2(
    kernel_name="google_bert_base_zh",  # "google_bert_base_en" for English
    labels=labels,
    device='cuda:0'
)

# Training the model
model.fit(
    sents_a_train=questions_train,
    sents_b_train=answers_train,
    labels_train=labels_train,
    model_saving_path='./fastbert.bin',
)

# Loading the model and making inference
model.load_model('./fastbert.bin')
label, exec_layers = model(
    sent_a='我也要用FastBERT!',
    sent_b='来，吃老干妈!',
    speed=0.7)

Usage

Args of FastBERT/FastBERT_S2:

Args	Type	Examples	Explanation
kernel_name	str	'google_bert_base_en'	The name of the kernel model, including 'google_bert_base_en', 'google_bert_base_zh'.
labels	list	['T', 'F']	A list of all labels.
seq_length (optional)	int	256	The sentence length for FastBERT. Default 128
device (optional)	str	'cuda:0'	The device for runing FastBERT, default 'cpu'

Args of FastBERT.fit():

Args	Type	Examples	Explanation
sentences_train	list	['sent 1', 'sent 2',...]	A list of training sentences.
labels_train	list	['T', 'F', ...]	A list of training labels.
batch_size (optional)	int	32	batch_size for training. Default 16
sentences_dev (optional)	list	[]	A list of validation sentences.
labels_dev (optional)	list	[]	A list of validation labels.
learning_rate (optional)	float	2e-5	learning rate.
finetuning_epochs_num (optional)	int	5	The epoch number of finetuning.
distilling_epochs_num (optional)	int	10	The epoch number of distilling.
report_steps (optional)	int	100	Report the training process every [report_steps] steps.
warmup (optional)	float	0.1	The warmup rate for training.
dev_speed (optional)	float	0.5	The speed for evaluating in the self-distilling process.
model_saving_path (optional)	str	'./model.bin'	The path to saving model.

Args of FastBERT.forward():

Args	Type	Examples	Explanation
sentence	str	'How are you'	The input sentence.
speed (optional)	float	0.5	The speed value for inference. Default 0.0.

Args of FastBERT_S2.fit():

Args	Type	Examples	Explanation
sents_a_train	list	['sent a 1', 'sent a 2',...]	A list of training A-sentences.
sents_b_train	list	['sent b 1', 'sent b 2',...]	A list of training B-sentences.
labels_train	list	['T', 'F', ...]	A list of training labels.
batch_size (optional)	int	32	batch_size for training. Default 16
sents_a_dev (optional)	list	[]	A list of validation A-sentences.
sents_b_dev (optional)	list	[]	A list of validation B-sentences.
labels_dev (optional)	list	[]	A list of validation labels.
learning_rate (optional)	float	2e-5	learning rate.
finetuning_epochs_num (optional)	int	5	The epoch number of finetuning.
distilling_epochs_num (optional)	int	10	The epoch number of distilling.
report_steps (optional)	int	100	Report the training process every [report_steps] steps.
warmup (optional)	float	0.1	The warmup rate for training.
dev_speed (optional)	float	0.5	The speed for evaluating in the self-distilling process.
model_saving_path (optional)	str	'./model.bin'	The path to saving model.

Args of FastBERT_S2.forward():

Args	Type	Examples	Explanation
sents_a	str	'How are you'	The input A-sentence.
sents_b	str	'How are you'	The input B-sentence.
speed (optional)	float	0.5	The speed value for inference. Default 0.0.