[Feat][BugFix]Support the Qwen3-Next-80B-A3B-Instruct quantization model&Fix the NZ issue (#4245)

### What this PR does / why we need it?
Support the Qwen3-Next-80B-A3B-Instruct quantization model and Fix the
NZ issue. Triton kernel doesn't support data format nz, thus we skip
converting weight to nz on layer `conv1d`

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: IncSec <1790766300@qq.com>

tests/ut/attention/test_mla_v1.py

@@ -797,7 +797,6 @@ class TestAscendMLAImpl(TestBase):
         self.assertEqual(q_pe.shape[1], self.impl.num_heads)
         self.assertEqual(q_pe.shape[2], self.impl.qk_rope_head_dim)
 
-    @patch('vllm_ascend.utils._ENABLE_NZ', True)
     @patch('torch_npu.npu_format_cast')
     def test_process_weights_after_loading(self, mock_format_cast):
         layer = MagicMock(spec=LinearBase)

tests/ut/models/test_qwen2_5_vl.py

@@ -1,5 +1,3 @@
-from unittest.mock import patch
-
 import pytest
 import torch
 import torch.nn.functional as F
@@ -367,7 +365,6 @@ class TestAscendQwen2_5_VisionTransformer(PytestBase):
         res = attention.pad_qkv_bias(torch.rand((300)))
         assert res.shape[0] == 384
 
-    @patch('vllm_ascend.utils._ENABLE_NZ', True)
     def test_pad_qkv_weight(self, mocker: MockerFixture):
         attention = self.init_vision_transformer(mocker)
         mocker.patch("torch.nn.Module.__setattr__")
@@ -380,7 +377,6 @@ class TestAscendQwen2_5_VisionTransformer(PytestBase):
         res = attention.pad_qkv_weight(torch.rand((300, 300)))
         assert res.shape == (384, 300)
 
-    @patch('vllm_ascend.utils._ENABLE_NZ', True)
     def test_pad_proj_weight(self, mocker: MockerFixture):
         attention = self.init_vision_transformer(mocker)
         mocker.patch("torch.nn.Module.__setattr__")

tests/ut/quantization/test_w4a8_dynamic.py

@@ -260,7 +260,6 @@ class TestAscendW4A8DynamicFusedMoEMethod(TestBase):
                 requires_grad=False)
         return layer
 
-    @patch('vllm_ascend.utils._ENABLE_NZ', False)
     @patch('torch_npu.npu_format_cast')
     @patch('torch_npu.npu_quantize')
     @patch('torch.Tensor.npu')

tests/ut/test_utils.py

@@ -46,18 +46,12 @@ class TestUtils(TestBase):
             self.assertFalse(utils.is_310p())
 
     def test_is_enable_nz(self):
-        # Case when _ENABLE_NZ is already set
-        utils._ENABLE_NZ = True
-        self.assertTrue(utils.is_enable_nz())
-
-        utils._ENABLE_NZ = False
-        self.assertFalse(utils.is_enable_nz())
-
-        # Case when _ENABLE_NZ is None and vllm_config is not provided
-        utils._ENABLE_NZ = None
-        with self.assertRaises(ValueError) as context:
-            utils.is_enable_nz()
-        self.assertIn("vllm_config must be provided", str(context.exception))
+        with mock.patch("vllm_ascend.utils.envs_ascend.VLLM_ASCEND_ENABLE_NZ",
+                        1):
+            self.assertTrue(utils.is_enable_nz())
+        with mock.patch("vllm_ascend.utils.envs_ascend.VLLM_ASCEND_ENABLE_NZ",
+                        0):
+            self.assertFalse(utils.is_enable_nz())
 
     def test_sleep_mode_enabled(self):
         utils._SLEEP_MODE_ENABLED = None

tests/ut/worker/test_worker_v1.py

@@ -281,9 +281,9 @@ class TestNPUWorker(TestBase):
 
             self.assertIn("Sleep mode is not enabled", str(cm.exception))
 
-    @patch('vllm_ascend.utils._ENABLE_NZ', False)
     @patch("vllm_ascend.worker.worker_v1.sleep_mode_enabled")
     @patch("vllm_ascend.worker.worker_v1.CaMemAllocator")
+    @patch.dict("os.environ", {"VLLM_ASCEND_ENABLE_NZ": "0"})
     def test_wake_up_mode_enabled(self, mock_allocator_class,
                                   mock_sleep_mode_enabled):
         """Test wake_up method when sleep mode is enabled"""