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 my understanding is correct, it seems that there are some error in Figure 1.
Since each model's parameter's delta is represented as (global diff) x (bit), it seems that the right side should be #models X (#parameters X 1bit + 16bit) instead of #parametersX (#models X 1bit + 16bit).
How do you think about this?
The text was updated successfully, but these errors were encountered:
Hi! Thank you for interesting idea!
If my understanding is correct, it seems that there are some error in Figure 1.
Since each model's parameter's delta is represented as (global diff) x (bit), it seems that the right side should be
#models X (#parameters X 1bit + 16bit)
instead of#parametersX (#models X 1bit + 16bit)
.How do you think about this?
The text was updated successfully, but these errors were encountered: