performance thoughts
#4
by
						
RonanMcGovern
	
							
						- opened
							
					
Thanks for making this model.
It's interesting how it seems weaker than Phi-2 - at least on coding. I notice there is an OpenHermes fine-tune too and it has the same issue (e.g. fails to write a function to add up the first N fibonacci numbers).
Any thoughts on why this might be the case?

