AutoTransformers Changelog

v1.2.0 (2024-04-24)

Added

• Support for Python 3.11

• Support for Torch 2.2.x

Changed

• Updated minimum Python version to 3.10

Removed

• Support for Python 3.7, 3.8 and 3.9

v1.1.0 (2023-06-05)

Added

• AutoTransformers can now be used to classify objects in images through the computer vision domain.

• AutoTransformers now support gradient clipping and gradient accumulation for training with FP16.

• Added MLM and PPB pre-training as well as fine-tuning tasks for documents.

• Added a new custom DeepOpinion Model: The DocumentModel (DocumentModelV4) that is more efficient and more accurate for Document Information Extraction as well as Document Classification.

Changed

• AutoTransformers now support the new OCRWrapper 1.0.0. Therefore, the data format changed.

• Subsampling was removed from the document domain.

• The default hyperparameters for documents were improved.

• Use safetensors to store checkpoints as well as models

v1.0.0 (2023-02-24)

The first version of AutoTransformers.