What are some good pre-trained models for factual/verifiable text generation and QA (without context)?
I’m referring to models that could be used for historical, medical, tech, research, and even wikipedia questions (without context) and text generation?
I’m having a hard time wrapping my head around this problem and I’d appreciate any insightful action-oriented input!
It would be awesome if it can be loaded with the api-inference…
I would start with T5 and try some mask filling/QA exercises to see if it works. Then try other models, also, you can check papers with code site to see which models are giving SOTA scores on the datasets you wish to use.
There are several models trained on wikipedia dataset Hugging Face – The AI community building the future.
Start with that and see if it gives good results,
There probably wont be one model+dataset that works for all your tasks, you will have to mix and match to see what works for you.