[fix] lint CI

Signed-off-by: anon189Ty <Stari_Falcon@outlook.com>

Author: anon189Ty

tests/ut/ops/test_linear.py

@@ -10,7 +10,6 @@
 from vllm_ascend.distributed import parallel_state
 from vllm_ascend.ops.linear import (AscendColumnParallelLinear,
                                     AscendMergedColumnParallelLinear,
-                                    AscendQKVParallelLinear,
                                     AscendRowParallelLinear,
                                     AscendUnquantizedLinearMethod)
 
@@ -179,64 +178,5 @@ def test_merged_mlp_tp_init(self):
         self.assertEqual(linear.forward_type, "mlp_tp")
 
 
-class TestAscendQKVParallelLinear(unittest.TestCase):
-
-    def setUp(self):
-        os.environ["VLLM_ASCEND_ENABLE_MLP_OPTIMIZE"] = "1"
-        # Mock get_tensor_model_parallel_world_size function
-        self.get_tensor_model_parallel_world_size_patch = mock.patch(
-            'vllm_ascend.ops.linear.get_tensor_model_parallel_world_size',
-            return_value=2)
-        self.get_tensor_model_parallel_world_size_mock = \
-            self.get_tensor_model_parallel_world_size_patch.start()
-        # Mock divide function (assumed to be in your module)
-        self.divide_patch = mock.patch('vllm_ascend.ops.linear.divide')
-        self.divide_mock = self.divide_patch.start()
-        self.divide_mock.side_effect = lambda x, y: x // y  # Simulate division
-
-        # Mock AscendMlpColumnParallelLinear's __init__
-        self.linear_init_patch = mock.patch.object(
-            AscendMlpColumnParallelLinear,
-            "__init__",
-            side_effect=self.mock_linear_init)
-        self.linear_init_patch.start()
-
-        # Create mock objects
-        self.quant_method_mock = mock.MagicMock()
-
-    def mock_linear_init(self, instance, *args, **kwargs):
-        torch.nn.Module.__init__(instance)
-        # Set quant_method and other attributes
-        instance.quant_method = self.quant_method_mock
-        instance.bias = torch.nn.Parameter(torch.randn(8))  # Example bias
-        instance.input_size = 16
-        instance.output_size = 8
-        instance.gather_output = False
-        instance.skip_bias_add = False
-        instance.return_bias = True
-
-    def testDown(self):
-        self.get_tensor_model_parallel_world_size_patch.stop()
-        self.divide_patch.stop()
-
-    def test_init_tpsize_larger_than_total_kv_head(self):
-        layer = AscendQKVParallelLinear(
-            hidden_size=16,
-            head_size=8,
-            total_num_heads=1,
-        )
-        self.assertEqual(layer.num_kv_heads, 1)
-        self.assertEqual(layer.num_kv_head_replicas, 2)
-
-    def test_init_tpsize_smaller_than_total_kv_head(self):
-        layer = AscendQKVParallelLinear(
-            hidden_size=16,
-            head_size=8,
-            total_num_heads=4,
-        )
-        self.assertEqual(layer.num_kv_heads, 2)
-        self.assertEqual(layer.num_kv_head_replicas, 1)
-
-
 if __name__ == '__main__':
     unittest.main()

vllm_ascend/ops/linear.py

@@ -23,8 +23,7 @@
 import torch_npu
 from torch.distributed import ProcessGroup
 from torch.nn.parameter import Parameter
-from vllm.distributed import (divide, get_tensor_model_parallel_world_size,
-                              split_tensor_along_last_dim)
+from vllm.distributed import divide, split_tensor_along_last_dim
 from vllm.distributed.parallel_state import get_tp_group
 from vllm.lora.utils import LinearBase
 from vllm.model_executor.layers.linear import (  # noqa