//
sign in
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
๐Ÿงต[2/n] ๐Ÿ’ก SFT Updates Are Dense ๐Ÿ’ก Unlike RL, Supervised Fine-Tuning (SFT) updates are much denser ๐Ÿง  ๐Ÿ“Š Sparsity is low โ€” at most only 15.31% of parameters remain untouched.
May 21, 2025
Sagnik Mukherjee