diff --git a/auto_round/inference/backend.py b/auto_round/inference/backend.py
index de9ea903f..0e824e609 100644
--- a/auto_round/inference/backend.py
+++ b/auto_round/inference/backend.py
@@ -456,7 +456,7 @@ def fp8_static_scheme_checker(
 
 BackendInfos["auto_round_kernel_xpu"] = BackendInfo(
     device=["xpu"],
-    sym=[True],
+    sym=[True, False],
     packing_format=GPTQ_FORMAT_NO_ZP,
     bits=[4, 8],
     group_size=None,
@@ -486,7 +486,7 @@ def fp8_static_scheme_checker(
 
 BackendInfos["auto_round_kernel_zp_xpu"] = BackendInfo(
     device=["xpu"],
-    sym=[True],
+    sym=[True, False],
     packing_format=GPTQ_FORMAT,
     bits=[4, 8],
     group_size=None,
@@ -500,7 +500,7 @@ def fp8_static_scheme_checker(
 )
 
 BackendInfos["auto_round_kernel_awq"] = BackendInfo(
-    device=["cpu"],
+    device=["cpu", "xpu"],
     sym=[True, False],
     packing_format=AWQ_FORMAT,
     bits=[4],
@@ -514,21 +514,6 @@ def fp8_static_scheme_checker(
     requirements=["torch>=2.8.0", "auto_round_kernel"],
 )
 
-BackendInfos["auto_round_kernel_awq_xpu"] = BackendInfo(
-    device=["xpu"],
-    sym=[True],
-    packing_format=AWQ_FORMAT,
-    bits=[4],
-    group_size=None,
-    priority=6,
-    checkers=[ark_feature_checker],
-    alias=["ark"],
-    compute_dtype=["float32", "float16"],
-    data_type=["int"],
-    act_bits=WOQ_DEFAULT_ACT_BITS,
-    requirements=["torch>=2.8.0", "auto_round_kernel"],
-)
-
 BackendInfos["ipex_gptq_cpu"] = BackendInfo(
     device=["cpu"],
     sym=[True, False],
diff --git a/auto_round_extension/ark/qlinear.py b/auto_round_extension/ark/qlinear.py
index 1e34ad4f6..a02f1643c 100644
--- a/auto_round_extension/ark/qlinear.py
+++ b/auto_round_extension/ark/qlinear.py
@@ -116,8 +116,7 @@ def post_init(self):
             raise NotImplementedError(
                 f"Device type {self.qweight.device.type} is not supported. Only CPU and XPU devices are supported."
             )
-        if self.qweight.device.type != "cpu" and self.asym:
-            raise NotImplementedError("Asymmetric quantization is only supported on CPU devices")
+
         if "awq" in self.QUANT_TYPE:
             intweight, zeros = unpack_awq(
                 self.qweight, self.qzeros, self.bits
diff --git a/test/test_ark/test_model.py b/test/test_ark/test_model.py
index 361f1bdf9..5fe1d09e3 100644
--- a/test/test_ark/test_model.py
+++ b/test/test_ark/test_model.py
@@ -29,8 +29,6 @@ def main_op(self, format, bits, group_size, sym, dtype, device, fast_cfg=True, t
             limit = 1000
             if not torch.xpu.is_available():
                 pytest.skip("No XPU device")
-            if sym is False:
-                pytest.skip("No asym support for XPU")
         model = AutoModelForCausalLM.from_pretrained(self.model_name, dtype="auto")
         tokenizer = AutoTokenizer.from_pretrained(self.model_name)
         if fast_cfg:
@@ -56,9 +54,9 @@ def main_op(self, format, bits, group_size, sym, dtype, device, fast_cfg=True, t
         shutil.rmtree(self.save_folder, ignore_errors=True)
 
     @pytest.mark.parametrize("format", ["auto_round", "auto_round:gptqmodel"])
-    @pytest.mark.parametrize("bits, group_size, sym", [(4, 128, True), (8, 128, True)])
+    @pytest.mark.parametrize("bits, group_size, sym", [(4, 128, False), (8, 128, True)])
     @pytest.mark.parametrize("dtype", [torch.bfloat16])
-    @pytest.mark.parametrize("device", ["cpu", "xpu"])
+    @pytest.mark.parametrize("device", ["xpu", "cpu"])
     def test_formats(self, format, bits, group_size, sym, dtype, device):
         self.main_op(format, bits, group_size, sym, dtype, device)