1 comment

[ 5.3 ms ] story [ 15.0 ms ] thread
Curious — how did you handle training stability early on? Was convergence an issue without heavy tuning?