v0.8.0

What's Changed

• Add interactive prepare_release.py by @ianbulovic in #224
• Modifications to allow for using ModernBERT as a base encoder. by @tmills in #226
• Add 9 publications using cnlp_transformers to README.md by @tmills in #231
• Train system and general refactoring by @ianbulovic in #230
• Unify model loading and REST APIs and improve CLI by @ianbulovic in #241

Full Changelog: v0.7.1...v0.8.0

v0.7.1

What's Changed

• Fix version by @tmills in #221

v0.7.0

What's Changed

• Allow continued training of hier models by @wonjininfo in #170
• Output prediction probs by @wonjininfo in #175
• Cleanup of several functions, adding doc strings and type annotations. by @tmills in #181
• added arguments for model selection metric by @spencerthomas1722 in #173
• Implement domain-adaptive pretraining by @angus-lherrou in #183
• Update drugs.com dataset URL by @mikix in #187
• docker: tweak docker builds, mostly to get gpus working by @mikix in #192
• Dev error analysis by @etgld in #172
• [FUNC] Add an option for truncation from left by @wonjininfo in #193
• Deleted import line for ALL_PRETRAINED_CONFIG_ARCHIVE_MAP (deprecated) by @guo0O0o in #197
• Experimental macOS/MPS support by @ianbulovic in #200
• [FIX] Updated structure_labels function to output softmax probability instead of logit for the best category by @wonjininfo in #204
• Add python 3.11, and update a bunch of other dependency versions. by @tmills in #207
• Error Analysis and Human Readable Predictions by @etgld in #206
• Dev tools by @ianbulovic in #208
• Lint, format, update README.md and CONTRIBUTING.md by @ianbulovic in #209
• Clean up api by @ianbulovic in #210
• Use uv for CI by @ianbulovic in #211
• Fix CI error from merge and speed up test collection by @ianbulovic in #212
• Fixes to data processing for updates to hf libraries. by @tmills in #214
• Remove punkt download. Fixes issue-213. by @tmills in #216
• Use setuptools-scm for versioning, update CONTRIBUTING.md by @ianbulovic in #217
• Consolidate CLI into cnlpt command by @ianbulovic in #218
• Removing references to rest api models which are currently broken. by @tmills in #219

New Contributors

• @mikix made their first contribution in #187
• @guo0O0o made their first contribution in #197
• @ianbulovic made their first contribution in #200

Full Changelog: v0.6.1...v0.7.0

v0.6.1

What's Changed

• Updated dependencies to fix huggingface hub issue in Docker builds

v0.6.0

What's Changed

• Training a single model can be done with multiple datasets and tasks, providing task names selects columns from datasets to train on.
• Move versioning into cnlpt.version by @angus-lherrou in #139
• Update CONTRIBUTING.md and GitHub Actions to reflect new dev branch naming convention by @angus-lherrou in #140
• merge docs and github actions changes into dev-v0.6.0 by @angus-lherrou in #144
• Realign baseline models with new task flow by @angus-lherrou in #150
• Improved Sphinx docs by @angus-lherrou in #154
  • Move type strings into annotations
  • Setup autodoc type hints
• Add support for loading local encoders for fine-tuning by @angus-lherrou in #152
• Drop support for Python 3.7 by @angus-lherrou in #161
• Add functionality to get chunk attentions from hierarchical model forward() so that we can visualize hierarchical attention. by @tmills in #162
• Add requirements for transform_prot by @angus-lherrou in #153
• Save config.json in training for CNN models; load best model before prediction (for all models) by @spencerthomas1722 in #164
• extract chemprot requirements from main setup.cfg (PyPI disallows direct dependency links) by @angus-lherrou in #168

New Contributors

• @spencerthomas1722 made their first contribution in #164

Full Changelog: v0.5.0...v0.6.0

v0.5.0

What's Changed

• Dockerization by @dogversioning in #108
• 103 add examples to readme for e2e relations and tagging using publicly available datasets by @etgld in #109
• Clarify build and release process in CONTRIBUTING.md by @angus-lherrou in #110
• Merge changes in main into v0.5.0 by @angus-lherrou in #113
• add early return with zero statuses by @angus-lherrou in #112
• Preloading models for docker servers by @dogversioning in #115
• Sync v0.5.0 with main by @angus-lherrou in #119
• Integrate changes to json format reading by @tmills in #117
• Move examples into subdirectory to clean up README by @vanh17 in #129
• Issue 130 - separate out hierarchical model to not inherit by @tmills in #131
• Update doctimerel rest to use current helper methods in cnlp_rest. by @tmills in #135
• Move hierarchical encoder head config into CnlpConfig by @angus-lherrou in #125

New Contributors

• @dogversioning made their first contribution in #108

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

The biggest change in this release is to the way datasets are managed. We now use Huggingface Datasets as a backend and so we have removed much of the dataset processing code. Fine-tuning on a new task no longer involves writing code, but it does require dataset to be in one of a few standard formats.

• v0.3.2: remove usage of deprecated ALL_PRETRAINED_CONFIG_ARCHIVE_MAP by @tmills in #98
• using new datasets library (motivated by mimic los task) by @tmills in #99
• V0.4.0 by @tmills in #104

Full Changelog: v0.3.2...v0.4.0

v0.3.2

What's Changed

• v0.3.2: remove usage of deprecated ALL_PRETRAINED_CONFIG_ARCHIVE_MAP by @angus-lherrou in #94

Full Changelog: v0.3.1...v0.3.2

v0.3.1

What's Changed

• Add GitHub Actions for building and testing new versions in #87
• Add basic testing capabilities in #89
• Bugfixes for #85, #88, and #91 in #93

Full Changelog: v0.3.0...v0.3.1

v0.3.0

The latest release contains bug fixes, better documentation, new features, and better information about how to contribute.