From 266ad0caca8b9c9308076bff3230143454a2380e Mon Sep 17 00:00:00 2001
From: Andrew White <andrewh@cdw.com>
Date: Mon, 27 Jul 2026 20:48:53 -0500
Subject: [PATCH] fix: Per-point LR adjustment is not persisted in optimizer
 state

---
 scene/per_point_adam.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/scene/per_point_adam.py b/scene/per_point_adam.py
index 6c99952..47db4bc 100644
--- a/scene/per_point_adam.py
+++ b/scene/per_point_adam.py
@@ -94,6 +94,7 @@ def step(self, closure=None):
                     scaled_step_size = step_size * per_point_lr
                     p.data.add_(-scaled_step_size * (exp_avg / denom))
                     per_point_lr = self._adjust_per_point_lr(per_point_lr, grad, mask)
+                    group['per_point_lr'] = per_point_lr
                 else:
                     p.data.addcdiv_(exp_avg, denom, value=-step_size)