fix(training): avoid double-counting MTP MoE metric layers (#5795)
Signed-off-by: Oded Ovadia <oded7272@gmail.com> Co-authored-by: gautham-kollu <gkollu@nvidia.com>
O
Oded Ovadia committed
8eea7a95e5e832c6c38f23fd991dfe5939baa469
Parent: d69c7c4
Committed by GitHub <noreply@github.com>
on 9/2/2026, 3:07:50 PM