test(mge/utils): skip network node convbias test in gpu

GitOrigin-RevId: 9d06cf6c5ad4c75840b5efb95278b8a7d989b73f

imperative/python/test/unit/utils/test_network_node.py

@@ -14,6 +14,7 @@ from megengine.core._imperative_rt.core2 import apply
 from megengine.core._wrap import Device
 from megengine.core.ops import builtin
 from megengine.device import is_cuda_available
+from megengine.distributed.helper import get_device_count_by_fork
 from megengine.functional.external import tensorrt_runtime_opr
 from megengine.jit.tracing import trace
 from megengine.tensor import Tensor
@@ -265,6 +266,10 @@ def test_deformable_ps_roi_pooling():
     check_pygraph_dump(fwd, [inp, rois, trans], [result])
 
 
+@pytest.mark.skipif(
+    get_device_count_by_fork("gpu") > 0,
+    reason="does not support int8 when gpu compute capability less than 6.1",
+)
 def test_convbias():
     @trace(symbolic=True, capture_as_const=True)
     def fwd(inp, weight, bias):