Model is developed in support of the University of Belgrade doctoral dissertation "Composite pseudogrammars based on parallel language models of Serbian" by Mihailo Škorić.
It generates semantically masked (lemmatized and without stopwords) sentences for Serbian.
This small gpt-2 model was fine-tuned on several corpora for Serbian (augmented using Latent semantic analysis methods).
The corpora include "The corpus of Contemporary Serbian", SrpELTeC and WikiKorpus by JeRTeh – Society for Language Resources and Technologies.
- Downloads last month