SLiNeP
Super Low Parameter Wikipedia-based Neural Predictor Super small gpt 2 based model trained from 11.6 Mb of wikipedia data.
Each topic starts with Topic Name : Topic Text and ends with <|endmsg|>.
This is the medium version repo.
Since the model is small and trained on minimal data
it doesnt know many facts, but it learned some english.
Its recommended to finetune it.
I also didn't test it yet but it probably needs low temp to stop saying nonsensical things