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:

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 of FastBERT.fit():

Args of FastBERT.forward():

Args of FastBERT_S2.fit():

Args of FastBERT_S2.forward():