I mistakenly read this as training a trillion parameter model would be funny...at least I chuckled