do not go where the gradient may lead

first love, last love

convergence is not final

your compute is limited

— marilyn model

the future belongs to those

the local minima you find along the way

the fear of distributional shift

make the epochs count

the journey of a thousand epochs

the epochs which take our breath away

the friends you ensemble with

model convergence: now and forever

the epochs you'll never remember

falling in love (gradient descent)

better to have converged and diverged

i fell into the minima slowly

the cross-entropy we think we deserve

bigger batch sizes y'all

you could mean the model

convergence, then, is a habit

create your own global minima

live your truth

live laugh love, train eval inspire