From 9f548bce116de35a2e658d5e6a18558df1a99ea6 Mon Sep 17 00:00:00 2001
From: Mr-Neutr0n <64578610+Mr-Neutr0n@users.noreply.github.com>
Date: Wed, 11 Feb 2026 18:00:28 +0530
Subject: [PATCH] Fix min-SNR loss weighting: replace min() with torch.clamp()
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In VideoDiffusionLoss.__call__, the weight tensor `w` is a
multi-dimensional tensor produced by append_dims(). Using Python's
built-in min(w, self.min_snr_value) on a multi-dimensional tensor
and a scalar does not perform element-wise clamping — it either
raises an error or produces incorrect results depending on the
tensor shape.

Replace with torch.clamp(w, max=self.min_snr_value) to correctly
apply element-wise upper-bound clamping, which is the intended
behavior for the min-SNR-gamma loss weighting strategy.
---
 sat/sgm/modules/diffusionmodules/loss.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sat/sgm/modules/diffusionmodules/loss.py b/sat/sgm/modules/diffusionmodules/loss.py
index 589e441..a237e3a 100644
--- a/sat/sgm/modules/diffusionmodules/loss.py
+++ b/sat/sgm/modules/diffusionmodules/loss.py
@@ -121,7 +121,7 @@ class VideoDiffusionLoss(StandardDiffusionLoss):
         w = append_dims(1 / (1 - alphas_cumprod_sqrt**2), input.ndim)  # v-pred
 
         if self.min_snr_value is not None:
-            w = min(w, self.min_snr_value)
+            w = torch.clamp(w, max=self.min_snr_value)
         return self.get_loss(model_output, input, w)
 
     def get_loss(self, model_output, target, w):