The discussions about SFT in pre training. If I understand correctly, the idea is that models that have pretraining data that contains some instruct data tend to learn to reason while those without any instruct data don't ever figure out how to reason.
Collin Stebbins
buildmine10
AI & ML interests
None yet
Recent Activity
commented on
an
article
26 days ago
Open R1: Update #2
new activity
11 months ago
microsoft/Phi-3-mini-128k-instruct:gguf
Organizations
None yet
buildmine10's activity
commented on
Open R1: Update #2
26 days ago
gguf
30
#24 opened 11 months ago
by
LaferriereJC