Sahajtomar/German_Zeroshot

Model has GBERT Large as base model and fine-tuned it on xnli de dataset . Default hypothesis template is in English: This text is {}. While using this model, change it to "In deisem geht es um {}" or something different. While inferencing through huggingface API may give poor results as it uses by default english template. Since model is monolingual and not multilingual, hypothesis template needs to be changed accordingly. This can be experimented with using a zero-shot-classification pipeline instead of a model that is more focused on German Zeroshot model . The default hypothesis template for this model is in German, but it can be,