From d1f93909ffcd0833c3f72862e3998886f0ca9e91 Mon Sep 17 00:00:00 2001
From: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
Date: Tue, 31 Mar 2026 21:48:47 +0000
Subject: [PATCH] clamp instead of use fp64

Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
---
 modelopt/torch/quantization/algorithms.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/modelopt/torch/quantization/algorithms.py b/modelopt/torch/quantization/algorithms.py
index 7907c79bd6..fe066293e9 100644
--- a/modelopt/torch/quantization/algorithms.py
+++ b/modelopt/torch/quantization/algorithms.py
@@ -764,7 +764,7 @@ def run_search(self):
 
 def _get_auto_quantize_score(grad_output, output_diff):
     x = grad_output.float() * output_diff.float()
-    return x.to(torch.float64).square().sum()
+    return x.clamp(-1e10, 1e10).square().sum()
 
 
 def _add_auto_quantize_score(grad_output, output_diff, score_tensor):