From 2885ec29ac41d685ffd2f9c146a8b18739a6fedf Mon Sep 17 00:00:00 2001
From: Benjamin Bennett <125957+benbennett@users.noreply.github.com>
Date: Tue, 23 Sep 2025 23:54:43 -0500
Subject: [PATCH] Detach loss tensors before scalar conversion

---
 models/base_model.py | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/models/base_model.py b/models/base_model.py
index d84ded8..a740b7d 100644
--- a/models/base_model.py
+++ b/models/base_model.py
@@ -123,11 +123,16 @@ def get_current_visuals(self):
 
     def get_current_losses(self):
         """Return traning losses / errors. simple.py will print out these errors on console, and save them to a file"""
-        errors_ret = OrderedDict()
-        for name in self.loss_names:
-            if isinstance(name, str):
-                errors_ret[name] = float(getattr(self, 'loss_' + name))  # float(...) works for both scalar tensor and float number
-        return errors_ret
+        errors_ret = OrderedDict()
+        for name in self.loss_names:
+            if isinstance(name, str):
+                loss_value = getattr(self, 'loss_' + name)
+                if isinstance(loss_value, torch.Tensor):
+                    # detach() avoids inadvertently keeping autograd history when converting tensors to scalars
+                    errors_ret[name] = loss_value.detach().item()
+                else:
+                    errors_ret[name] = float(loss_value)
+        return errors_ret
 
     def save_networks(self, epoch):
         """Save all the networks to the disk.