Changed - Update dependencies with `torchoutil`, and clean a lot of dead code. - Convnext model (BL version) is now automatically selected during HDF preparation. ([5](https://github.com/Labbeti/conette-audio-captioning/issues/5))
Fixed - Requirements versions specified during installation. ([4](https://github.com/Labbeti/conette-audio-captioning/issues/4)) - Preparation now correctly loads `scale_layers` in ConvNext during preparation. ([5](https://github.com/Labbeti/conette-audio-captioning/issues/5)) - ConvNeXt 75 iterations is now loaded instead of ConvNeXt 70 iterations for CoNeTTE.
0.2.2
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.
0.2.1
Added - `conette-predict` now support CNext-trans (baseline) model.
0.2.0
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.
0.1.4
Fixed - Fix forbid repetition mode argument.
0.1.3
Added - Forbid repetition mode argument to LightningModule and HuggingFace wrapper.