Shows an implementation of model soups (https://arxiv.org/abs/2203.05482) for text classification models.
APACHE-2.0 License
Shows an implementation of model soups [1] for text classification models.
notebooks/train-models.ipynb
shows how to obtain the fine-tuned text classificationTrainer
and AutoModelForSequenceClassification
classes from Transformers.notebooks/model-soups.ipynb
shows how to obtain model soups using two recipes presented in [1].notebooks/logit-ensembling.ipynb
shows how to perform the classic logit ensembling. It's here forTBA
[1] Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time (ICML 2022)