Code2Seq | 1.2.0

• Update dependencies for PyTorch and PyTorch-Lightning
• Use Rich progress bar while training
• Report ChrF metric for validation and test holdouts for code2seq model

Code2Seq | 1.1.1

• Vocabulary is built based on the number of occurrences of each token (used top-k before).
• Decoder now returns attention weights
• More flexible wandb configuration
• Bugs fixes

Fix reseting metrics

Call reset for each metric after end of epoch

Improve uploading callback behaviour

Update uploading checkpoint behaviour with respect to the new version of commode-utils

Optimize dataset usage

Remove using torch tensors from the path context dataset. Now, it returns lists of ints. This should speed up calculations.

code2seq v.1.0

• Integrate commode-utils
• Prettify all pipelines
• Add tests
• fix bugs

Specify install requirements

Remove dev dependencies from installation requirements. For now, black or mypy will not be installed along with code2seq.

Update astminer dependency

0.0.3

Update version number

Correct CI&CD pipeline

Fixing pipeline in order to publish code2seq to the pip after each release.

code2seq release

0.0.1

Publish to pypi in case of releasing