PyTorch ignite
- Refactored code to move the data dispatch step to base class - reduce a lot of repeated steps in derived logger classes
- Added feature to log state attributes along with metrics in Loggers.
- Combined State Attributes with Metrics so that they can be handled together.
- Added ROUGE metric for batch calculation.
- Added the state attributes and corresponding tests for
- Tensorboard
- Polyaxon
- Tqdm
- ClearML
- WandB
- MLFlow
- Visdon
Blogs
- Tutorial for Multi GPU/TPU seq2seq models using Ignite. \href{https://pytorch-ignite.ai/tutorials/intermediate/02-machine_translation_using_pytorch_ignite/}{[Ignite Website]}}
- Continuous Diffusion Models for NLP \href{https://medium.com/p/e7a464c1da4}{[Medium]}}
- Skin Segmentation in Images \href{https://ishankumar216.github.io/2021-03-07-Skin-Color-Segmentation-Methods/}{[Website]}}
- BART and MBART \href{https://ishankumar216.github.io/2021-03-07-BART-and-MBART/}{[Website]}}
PyTorch vision
- Ported affine transform tests from unittest to pytest.
- Ported grayscale transform tests from unittest to pytest