Skip to main content

a crf layer for tensorflow 2 keras

Project description


  • a simple CRF layer for tensorflow 2 keras
  • support keras masking


$ pip install tf2crf


  • easy to use CRF layer with tensorflow
  • support mixed precision training
  • support the ModelWithCRFLossDSCLoss with DSC loss, which increases f1 score with unbalanced data (refer the paper Dice Loss for Data-imbalanced NLP Tasks)


  • Add internal kernel like CRF in keras_contrib, so now there is no need to stack a Dense layer before the CRF layer.
  • I have changed the previous way that putting loss function and accuracy function in the CRF layer. Instead I choose to use ModelWappers (refered to jaspersjsun), which is more clean and flexible.


tensorflow >= 2.1.0 Recommmend use the latest tensorflow-addons which is compatiable with your tf version.


import tensorflow as tf
from tf2CRF import CRF
from tensorflow.keras.layers import Input, Embedding, Bidirectional, GRU, Dense
from tensorflow.keras.models import Model
from tf2crf import CRF, ModelWithCRFLoss

inputs = Input(shape=(None,), dtype='int32')
output = Embedding(100, 40, trainable=True, mask_zero=True)(inputs)
output = Bidirectional(GRU(64, return_sequences=True))(output)
crf = CRF(units=9, type='float32')
output = crf(output)
base_model = Model(inputs, output)
model = ModelWithCRFLoss(base_model, sparse_target=True)

x = [[5, 2, 3] * 3] * 10
y = [[1, 2, 3] * 3] * 10, y=y, epochs=2, batch_size=2)'tests/1')

Project details

Download files

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

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

tf2crf-0.1.33-py2.py3-none-any.whl (7.3 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page