MindNLP

News 📢

🔥 Latest Features
- 📃 Support PreTrained Models, including BERT, Roberta, GPT2 and T5. You can use them by following code snippet:
```
from mindnlp.models import BertModel

model = BertModel.from_pretrained('bert-base-cased')
```

Introduction

MindNLP is an open source NLP library based on MindSpore. It supports a platform for solving natural language processing tasks, containing many common approaches in NLP. It can help researchers and developers to construct and train models more conveniently and rapidly.

The master branch works with MindSpore master.

Major Features

Comprehensive data processing: Several classical NLP datasets are packaged into friendly module for easy use, such as Multi30k, SQuAD, CoNLL, etc.
Friendly NLP model toolset: MindNLP provides various configurable components. It is friendly to customize models using MindNLP.
Easy-to-use engine: MindNLP simplified complicated training process in MindSpore. It supports Trainer and Evaluator interfaces to train and evaluate models easily.

Quick Links

Installation

Dependency

mindspore >= 1.8.1

Install from source

To install MindNLP from source, please run:

pip install git+https://github.com/mindspore-lab/mindnlp.git

Get Started

We will next quickly implement a sentiment classification task by using mindnlp.

Define Model

from mindspore import ops
from mindnlp.abc import Seq2vecModel

class SentimentClassification(Seq2vecModel):
    def construct(self, text):
        _, (hidden, _), _ = self.encoder(text)
        context = ops.concat((hidden[-2, :, :], hidden[-1, :, :]), axis=1)
        output = self.head(context)
        return output

Define Hyperparameters

The following are some of the required hyperparameters in the model training process.

# define Models & Loss & Optimizer
hidden_size = 256
output_size = 1
num_layers = 2
bidirectional = True
drop = 0.5
lr = 0.001

Data Preprocessing

The dataset was downloaded and preprocessed by calling the interface of dataset in mindnlp.

Load dataset:

from mindnlp import load_dataset

imdb_train, imdb_test = load_dataset('imdb', shuffle=True)

Initializes the vocab and tokenizer for preprocessing:

from mindnlp import Vocab
from mindnlp.transforms import BasicTokenizer

tokenizer = BasicTokenizer(True)
vocab = Vocab.from_pretrained(name="glove.6B.100d")

The loaded dataset is preprocessed and divided into training and validation:

from mindnlp.dataset import process

imdb_train = process('imdb', imdb_train, tokenizer=tokenizer, vocab=vocab, \
                     bucket_boundaries=[400, 500], max_len=600, drop_remainder=True)
imdb_test = process('imdb', imdb_test, tokenizer=tokenizer, vocab=vocab, \
                     bucket_boundaries=[400, 500], max_len=600, drop_remainder=False)

Instantiate Model

from mindnlp.modules import RNNEncoder, Glove

embedding = Glove.from_pretrained('6B', 100, special_tokens=["<unk>", "<pad>"])
# build encoder
lstm_layer = nn.LSTM(100, hidden_size, num_layers=num_layers, batch_first=True,
                     dropout=dropout, bidirectional=bidirectional)
encoder = RNNEncoder(embedding, lstm_layer)

# build head
head = nn.SequentialCell([
    nn.Dropout(p=dropout),
    nn.Sigmoid(),
    nn.Dense(hidden_size * 2, output_size,
             weight_init=HeUniform(math.sqrt(5)),
             bias_init=Uniform(1 / math.sqrt(hidden_size * 2)))

])

# build network
network = SentimentClassification(encoder, head)
loss = nn.BCELoss(reduction='mean')
optimizer = nn.Adam(network.trainable_params(), learning_rate=lr)

Training Process

Now that we have completed all the preparations, we can begin to train the model.

from mindnlp.engine.metrics import Accuracy
from mindnlp.engine.trainer import Trainer

# define metrics
metric = Accuracy()

# define trainer
trainer = Trainer(network=network, train_dataset=imdb_train, eval_dataset=imdb_test, metrics=metric,
                  epochs=5, loss_fn=loss, optimizer=optimizer)
trainer.run(tgt_columns="label")

License

This project is released under the Apache 2.0 license.

Feedbacks and Contact

The dynamic version is still under development, if you find any issue or have an idea on new features, please don't hesitate to contact us via Github Issues.

Acknowledgement

MindSpore is an open source project that welcome any contribution and feedback.
We wish that the toolbox and benchmark could serve the growing research
community by providing a flexible as well as standardized toolkit to reimplement existing methods
and develop their own new semantic segmentation methods.

Citation

If you find this project useful in your research, please consider citing:

@misc{mindnlp2022,
    title={{MindNLP}: a MindSpore NLP library},
    author={MindNLP Contributors},
    howpublished = {\url{https://github.com/mindlab-ai/mindnlp}},
    year={2022}
}

linorman / mindnlp Goto Github PK

mindnlp's Introduction