This model was mostly created as a comparative benchmark for another model, see here: https://huggingface.co/MoritzLaurer/deberta-v3-base-zeroshot-v1.1-all-33

This model was only trained on five NLI datasets, while the other model was trained on many more datasets. I mostly recommend using the other model. This NLI-only model might only be better for tasks that are not zeroshot classification but that adhere more strictly to the original NLI task.

See the other model's model card for usage instructions, training data and the paper.