Resource-friendly CPU version with enhanced diff computing and Docker image

This release marks an important milestone in heptabot development, as all of our major goals have now been reached. The project was split into two branches, cpu with support for lightweight PyTorch model that runs only on CPU and gpu-tpu with more sophisticated T5 v. 1.0 TensorFlow models capable of producing better results. We also release three versions, tiny, medium and xxl, which differ by the checkpoint size of T5 text generation model employed. For each of these versions, we provide a notebook so that they could be conveniently used in Google Colab or Kaggle Kernels. This release also features updated installation procedures (we switched to Docker images as they are more stable and more convenient to work with), enhanced diff computing between the original and corrected texts (the error spans now align better), a function to convert heptabot's output to .ann format used by brat and some bug fixes for better stability.

Enhanced Colab notebook, new (very fast) TPU interface, enhanced diff computing and Docker image

This release marks an important milestone in heptabot development, as all of our major goals have now been reached. The project was split into two branches, cpu with support for lightweight PyTorch model that runs only on CPU and gpu-tpu with more sophisticated T5 v. 1.0 TensorFlow models capable of producing better results. We also release three versions, tiny, medium and xxl, which differ by the checkpoint size of T5 text generation model employed. For each of these versions, we provide a notebook so that they could be conveniently used in Google Colab or Kaggle Kernels. This release also features updated installation procedures (we switched to Docker images as they are more stable and more convenient to work with), enhanced diff computing between the original and corrected texts (the error spans now align better), a function to convert heptabot's output to .ann format used by brat and some bug fixes for better stability.

1.0 – Finalized development of T5 v1-based model

This release marks an important point in the development of Tensorflow-based heptabot implementation, as for now all the files have been pushed and all the known issues have been resolved. We are interested to implement syntetic training data and introduce fixes to retraining procedure in the future updates; we also, however, are interested in developing a version which could be served from CPU, and these pytorch-based implementations will be moved to 2.0 branch.

0.1.1 – First working build

This just works (c). First release with all things working as intended