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
As far as I understand, right now text pairs are picked directly using Datapack samples. However it is almost impossible to mine "hard" negative samples before training. Just adding into class 0/ rank 0.0 random pairs which are not corresponding is very wasteful and inefficient.
The text was updated successfully, but these errors were encountered:
As far as I understand, right now text pairs are picked directly using Datapack samples. However it is almost impossible to mine "hard" negative samples before training. Just adding into class 0/ rank 0.0 random pairs which are not corresponding is very wasteful and inefficient.
The text was updated successfully, but these errors were encountered: