You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
if pp = 0 (and the model is initiated using megatron's base model instead of the pipeline parallel model) then it considers a single microbatch a step, and so the flops/timer logging is inaccurate.
The text was updated successfully, but these errors were encountered:
if pp = 0 (and the model is initiated using megatron's base model instead of the pipeline parallel model) then it considers a single microbatch a step, and so the flops/timer logging is inaccurate.
The text was updated successfully, but these errors were encountered: