update paper citation for act mem (#59)

flop calculation fixes (#54)

add muon and sgd_momentum optimizers (#55)

add non-int ffn expansion factor (#53)

add alltoallv support for MPI alltoall bench (#51)

add reduce scatter communication benchmark (#50)

add scientific notation, fix token arg dtype, and clean up help msg (#49)

fix layernorm param calc (#48)

add mup (#47)

add name to transformer inference arithmetic (#46)

Add HuggingFace arg so that arch is automatic (#39)

--swiglu conflicts with --num-mlp-linears (#44)

Add transformer explainer

cleanup

remove duplicate ffn expansion factor arg from flops calc