Comment by domenicrosati

The T0 series by big science is essentially an instruct model (though using multitask prompting instead of user feedback). You should check it out. I have got very competitive results on prompting t0-11b v instructgpt3(text davinci 2)