diff --git a/python/paddle/nn/quant/quantized_linear.py b/python/paddle/nn/quant/quantized_linear.py
index 56058f0fa8dc3c..868df9711313bb 100644
--- a/python/paddle/nn/quant/quantized_linear.py
+++ b/python/paddle/nn/quant/quantized_linear.py
@@ -21,7 +21,6 @@
 from paddle.base.data_feeder import check_dtype
 from paddle.device import (
     is_compiled_with_cuda,
-    is_compiled_with_rocm,
 )
 from paddle.device.cuda import get_device_capability
 from paddle.framework import (
@@ -43,7 +42,7 @@
 
 def _get_arch_info():
     # Get SMVersion from device.
-    if is_compiled_with_cuda() or is_compiled_with_rocm():
+    if is_compiled_with_cuda():
         cuda_version = paddle.version.cuda()
         if (
             cuda_version is not None and cuda_version != 'False'