Version 0.4.0

[0.4.0] 2025-11-06

Changed

• BREAKING: Constraint python version to be 3.10 to avoid errors with newer pytorch/huggingface packages.

Version 0.3.2

[0.3.2] 2024-12-13

Added

• Add MIT LICENCE file.

Changed

• Update README.md and CITATION.cff file.

Version 0.3.1

[0.3.1] 2024-06-10

Fixed

• Remove duplicated files in results folder.
• CNN10 and CNN14D pre-process transform instantiation.

Version 0.3.0

[0.3.0] 2024-04-18

Changed

• Update dependencies with torchoutil, and clean a lot of dead code.
• Convnext model (BL version) is now automatically selected during HDF preparation. (#5)

Fixed

• Requirements versions specified during installation. (#4)
• Preparation now correctly loads scale_layers in ConvNext during preparation. (#5)
• ConvNeXt 75 iterations is now loaded instead of ConvNeXt 70 iterations for CoNeTTE.

Version 0.2.2

[0.2.2] 2024-01-15

Added

• Multiple candidates, predictions and probabilities in model outputs.
• train_and_enable_grad method in CoNeTTEModel class.

Changed

• Rename eval_and_detach to eval_and_disable_grad in CoNeTTEModel class.

Version 0.2.1

[0.2.1] 2024-01-12

Added

• conette-predict now support CNext-trans (baseline) model.

Version 0.2.0

[0.2.0] 2024-01-12

Added

• CoNeTTE training source code, with entire data processing.
• ConvNeXt-trans baseline training source code, with entire data processing.
• ConvNeXt tag logits to CoNeTTE model outputs during inference.

Version 0.1.4

[0.1.4] 2023-11-20

Fixed

• Fix forbid repetition mode argument.

Version 0.1.3

[0.1.3] 2023-11-20

Added

• Forbid repetition mode argument to LightningModule and HuggingFace wrapper.

Version 0.1.2

[0.1.2] 2023-11-17

Fixed

• Task embeddings inputs wavcaps_audioset_sl and wavcaps_bbc_sound_effects.