search

humanloop / memorai

🧠 AI memory assistant – remember everything you read
https://memorai.humanloop.ml
295 stars 20 forks source link

Change the ml model to valhalla/t5-base-qg #16

Open McSpooder opened 3 years ago

McSpooder commented 3 years ago

This is less of an issue and more of a feature suggestion.

So instead of picking a set of words to generate cloze sentences, generate contextual questions and answers. This can be done using the valhalla/t5-base-qg-hl model from hugging face. (https://huggingface.co/valhalla/t5-base-qg-hl?text=%3Chl%3EPeople%3Chl%3E+celebrate+Christmas+on+25th+of+December.)

On the site a proper question can be generate from an input string where the answer is enclosed within tags. So whatever is being used for the cloze word selection should be used for the tags enclosure. Then each separate string should be ran through the model to generate multiple questions.

PeadarOhAodha commented 3 years ago

Thanks @McSpooder - we've wanted to add some more sophisticated question gen techniques into the backend if people would find that useful... I'll take a look at your suggestion in the next few days.