Skip to main content

Transformer-XL implemented in Keras

Project description

Keras Transformer-XL

Travis Coverage Version Downloads License

[中文|English]

Unofficial implementation of Transformer-XL.

Install

pip install keras-transformer-xl

Usage

Load Pretrained Weights

Several configuration files can be found at the info directory.

import os
from keras_transformer_xl import load_trained_model_from_checkpoint

checkpoint_path = 'foo/bar/sota/enwiki8'
model = load_trained_model_from_checkpoint(
    config_path=os.path.join(checkpoint_path, 'config.json'),
    checkpoint_path=os.path.join(checkpoint_path, 'model.ckpt')
)
model.summary()

About IO

The generated model has two inputs, and the second input is the lengths of memories.

You can use MemorySequence wrapper for training and prediction:

import keras
import numpy as np
from keras_transformer_xl import MemorySequence, build_transformer_xl


class DummySequence(keras.utils.Sequence):

    def __init__(self):
        pass

    def __len__(self):
        return 10

    def __getitem__(self, index):
        return np.ones((3, 5 * (index + 1))), np.ones((3, 5 * (index + 1), 3))


model = build_transformer_xl(
    units=4,
    embed_dim=4,
    hidden_dim=4,
    num_token=3,
    num_block=3,
    num_head=2,
    batch_size=3,
    memory_len=20,
    target_len=10,
)
seq = MemorySequence(
    model=model,
    sequence=DummySequence(),
    target_len=10,
)

model.predict(model, seq, verbose=True)

Use tf.keras

Add TF_KERAS=1 to environment variables to use tensorflow.python.keras.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for keras-transformer-xl, version 0.11.0
Filename, size File type Python version Upload date Hashes
Filename, size keras-transformer-xl-0.11.0.tar.gz (15.7 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page