[bugfix]change log2phy map to npu (#3339)

### What this PR does / why we need it?
Resolved the issue of EPLB failure caused by changes in the log2phy map
due to device type modifications when using MTP rotation position
encoding.

### Does this PR introduce any user-facing change?

### How was this patch tested?
https://github.com/vllm-project/vllm/commit/releases/v0.11.0


- vLLM version: v0.11.0

---------

Signed-off-by: offline0806 <3337230449@qq.com>
Co-authored-by: offline0806 <3337230449@qq.com>

tests/ut/torchair/ops/test_torchair_fused_moe.py

@@ -56,7 +56,8 @@ def mock_dist_env(mocker: MockerFixture):
     # init dist env patch
     dp_metadata = MagicMock(num_tokens_across_dp_cpu=[5, 5])
 
-    with patch('torch.distributed.get_rank', return_value=0), \
+    with patch('torch.npu.is_available', return_value=True), \
+         patch('torch.distributed.get_rank', return_value=0), \
          patch('torch.distributed.get_world_size', return_value=4), \
          patch('vllm_ascend.torchair.ops.torchair_fused_moe.get_ep_group', return_value=mock_ep_and_mc2_group(mocker)), \
          patch('vllm_ascend.torchair.ops.torchair_fused_moe.get_mc2_group', return_value=mock_ep_and_mc2_group(mocker)), \
@@ -208,6 +209,7 @@ class MockFusedMoEMethod(FusedMoEMethodBase):
 
 class TestTorchairAscendFusedMoe:
 
+    @pytest.fixture
     def test_init_no_quant(self, mock_dist_env, default_moe_config):
         layer = TorchairAscendFusedMoE(**default_moe_config)
 
@@ -237,6 +239,7 @@ class TestTorchairAscendFusedMoe:
             error_config['scoring_func'] = "random"
             layer = TorchairAscendFusedMoE(**error_config)
 
+    @pytest.fixture
     def test_init_with_quant(self, mock_dist_env, default_moe_config):
         mock_quant_config = MagicMock()
         mock_quant_method = MockFusedMoEMethod()
@@ -248,6 +251,7 @@ class TestTorchairAscendFusedMoe:
             assert moe.quant_method is not None
             assert isinstance(moe.quant_method, AscendFusedMoEMethod)
 
+    @pytest.fixture
     def test_init_with_mixed_quant(self, mock_dist_env, default_moe_config):
         mock_quant_config = MagicMock()
         mock_quant_method = MockFusedMoEMethod()
@@ -261,6 +265,7 @@ class TestTorchairAscendFusedMoe:
         assert isinstance(moe.quant_method,
                           TorchairAscendUnquantizedFusedMoEMethod)
 
+    @pytest.fixture
     @pytest.mark.parametrize(
         "others_param",
         [[None,
@@ -306,6 +311,7 @@ class TestTorchairAscendFusedMoe:
         else:
             assert output.shape == (num_tokens, 32)
 
+    @pytest.fixture
     def test_forward_ms_fused_moe_comp(self, mock_dist_env,
                                        default_moe_config):
         inputs = torch.randn(5, 32)

vllm_ascend/ops/common_fused_moe.py

@@ -204,7 +204,7 @@ class AscendFusedMoE(FusedMoE):
                     self.global_redundant_expert_num)
                 self.log2phy = determine_default_log2phy_map(
                     self.global_num_experts, self.ep_size, self.ep_rank,
-                    self.global_redundant_expert_num)
+                    self.global_redundant_expert_num).npu()
         local_num_experts = (torch.sum(
             self.expert_map != -1) if self.expert_map is not None else
                              self.global_num_experts)

vllm_ascend/torchair/ops/torchair_fused_moe.py

@@ -1045,7 +1045,7 @@ class TorchairAscendFusedMoE(FusedMoE):
                     self.global_redundant_expert_num)
                 self.log2phy = determine_default_log2phy_map(
                     self.global_num_experts, self.ep_size, self.ep_rank,
-                    self.global_redundant_expert_num)
+                    self.global_redundant_expert_num).npu()
         local_num_experts = (torch.sum(self.expert_map != -1)
                              if self.expert_map is not None else num_experts)
         if self.dynamic_eplb: