Try gradient loss in Samudra
#525 · Python · ★ 26 · Apache-2.0 · updated 22d ago
Might want to do #497 or enable dynamic MSE + gradients first. Train Train: experiment 3 Intended board status: Backlog. Compare a horizontal-gradient loss against the baseline resulting from experiment 2. Scheduling is blocked on that experiment's report and accept/reject…
replies in ~22hnobody circling
Cleared 1m ago