test ut fixup

Signed-off-by: CodeNine-CJ <[email protected]>

Author: ForBetterCodeNine

tests/ut/torchair/test_torchair_model_runner.py

@@ -65,75 +65,76 @@ class TestNPUTorchairModelRunner(PytestBase):
     #     return runner
     @pytest.fixture
     def setup_npu_torchair_model_runner(self, mocker: MockerFixture):
-        # 核心配置对象
         vllm_config = MagicMock(spec=VllmConfig)
-        
-        # --- 必需的配置对象 (用于初始化属性) ---
         vllm_config.model_config = MagicMock()
+        cache_config = CacheConfig(block_size=16)
+        vllm_config.cache_config = cache_config
         vllm_config.model_config.hf_config = MagicMock()
         vllm_config.model_config.hf_config.index_topk = 2
-        vllm_config.model_config.max_model_len = 1024 # 模拟 model_config.max_model_len
-        vllm_config.model_config.use_mla = False # 模拟 model_config.use_mla
-        vllm_config.model_config.get_hidden_size.return_value = 512 # 模拟 get_hidden_size
-        vllm_config.model_config.pooler_config = None # 模拟 is_pooling_model
-        vllm_config.model_config.logits_processors = [] # 模拟 build_logitsprocs
-
+        vllm_config.model_config.max_model_len = 1024
+        vllm_config.model_config.use_mla = False
+        vllm_config.model_config.get_hidden_size.return_value = 512
+        vllm_config.model_config.pooler_config = None
+        vllm_config.model_config.logits_processors = []
         cache_config = MagicMock()
         cache_config.block_size = 16
-        cache_config.cache_dtype = "auto" # 模拟 cache_config.cache_dtype
+        cache_config.cache_dtype = "auto"
         vllm_config.cache_config = cache_config
-        
+
         speculative_config = MagicMock()
         speculative_config.num_speculative_tokens = 4
         vllm_config.speculative_config = speculative_config
-        
+
         vllm_config.compilation_config = MagicMock()
-        vllm_config.compilation_config.cudagraph_mode = Mock() # 模拟 compilation_config
-        # 模拟 aclgraph_batch_sizes
-        vllm_config.compilation_config.cudagraph_capture_sizes = [1, 2, 4] 
-        
+        vllm_config.compilation_config.cudagraph_mode = Mock()
+        vllm_config.compilation_config.cudagraph_capture_sizes = [1, 2, 4]
+
         vllm_config.lora_config = MagicMock()
         vllm_config.parallel_config = MagicMock()
-        vllm_config.parallel_config.data_parallel_size = 1 # 模拟 dp_size
-        vllm_config.parallel_config.data_parallel_rank = 0 # 模拟 dp_rank
-        vllm_config.parallel_config.cp_kv_cache_interleave_size = 1 # 模拟 cp_kv_cache_interleave_size
+        vllm_config.parallel_config.data_parallel_size = 1
+        vllm_config.parallel_config.data_parallel_rank = 0
+        vllm_config.parallel_config.cp_kv_cache_interleave_size = 1
 
         scheduler_config = MagicMock()
-        scheduler_config.max_num_batched_tokens = 2048 # 模拟 max_num_tokens
-        scheduler_config.max_num_seqs = 64 # 模拟 max_num_reqs (decode_max_num_seqs 默认为 0)
-        scheduler_config.chunked_prefill_enabled = True # 模拟 chunked_prefill_enabled
-        scheduler_config.async_scheduling = False # 模拟 use_async_scheduling
+        scheduler_config.max_num_batched_tokens = 2048
+        scheduler_config.max_num_seqs = 64
+        scheduler_config.chunked_prefill_enabled = True
+        scheduler_config.async_scheduling = False
         vllm_config.scheduler_config = scheduler_config
 
-        # --- 修复 'load_config' 报错 ---
-        vllm_config.load_config = MagicMock() 
+        vllm_config.load_config = MagicMock()
 
-        # --- 模拟 kv_transfer_config (用于判断 kv role) ---
         vllm_config.kv_transfer_config = None
 
-        # --- 模拟其他缺失的函数/常量 ---
-        mocker.patch("vllm_ascend.worker.model_runner_v1.is_pin_memory_available",
-                     return_value=True) # 模拟 pin_memory
+        mocker.patch(
+            "vllm_ascend.worker.model_runner_v1.is_pin_memory_available",
+            return_value=True)
         mocker.patch("vllm_ascend.worker.model_runner_v1.cdiv",
-                     return_value=64) # 模拟 max_num_blocks_per_req
-        mocker.patch("vllm_ascend.worker.model_runner_v1.prefill_context_parallel_enable",
-                     return_value=False) # 模拟 pcp_size/rank
-        mocker.patch("vllm_ascend.worker.model_runner_v1.get_dcp_group").return_value.world_size = 1 # 模拟 dcp_size/rank
-        mocker.patch("vllm_ascend.torchair.torchair_model_runner.get_attn_backend",
-                     autospec=True) # 模拟 Attention 设置
-        mocker.patch("vllm_ascend.torchair.torchair_model_runner._set_up_drafter") # 模拟 drafter 设置
-        mocker.patch("vllm_ascend.torchair.torchair_model_runner._may_pad_kv_consumer_num_seq") # 模拟 kv 填充
-
-        # NPU特定的配置
+                     return_value=64)
+        mocker.patch(
+            "vllm_ascend.worker.model_runner_v1.prefill_context_parallel_enable",
+            return_value=False)
+        mocker.patch("vllm_ascend.worker.model_runner_v1.get_dcp_group"
+                     ).return_value.world_size = 1
+        mocker.patch(
+            "vllm_ascend.torchair.torchair_model_runner.get_attn_backend",
+            autospec=True)
+        mocker.patch(
+            "vllm_ascend.torchair.torchair_model_runner._set_up_drafter")
+        mocker.patch(
+            "vllm_ascend.torchair.torchair_model_runner._may_pad_kv_consumer_num_seq"
+        )
+
+
         device = torch.device("npu:0")
         ascend_config = MagicMock()
-        # 确保 ascend_scheduler_config.enabled 被设置，否则 chunked_prefill_enabled 会被设置为 True
-        ascend_config.ascend_scheduler_config.enabled = False 
-        # 其他 NPU/Ascend 特有配置
-        ascend_config.weight_prefetch_config = Mock() 
+
+        ascend_config.ascend_scheduler_config.enabled = False
+
+        ascend_config.weight_prefetch_config = Mock()
         ascend_config.dynamic_eplb = False
         ascend_config.expert_map_record_path = None
-        
+
         mocker.patch("vllm_ascend.utils.get_ascend_config",
                      return_value=ascend_config)
         mocker.patch("vllm_ascend.torchair.utils.register_torchair_model")
@@ -146,9 +147,8 @@ def setup_npu_torchair_model_runner(self, mocker: MockerFixture):
         mock_attn_backend = Mock()
         mock_attn_backend.get_builder_cls.return_value = lambda *args, **kwargs: mock_attn_builder
 
-        # 设置类属性（如果需要）
         NPUTorchairModelRunner.decode_token_per_req = 1
-        NPUTorchairModelRunner.max_num_tokens = 10 # 这行在实际测试中可能被覆盖
+        NPUTorchairModelRunner.max_num_tokens = 10
 
         runner = NPUTorchairModelRunner(vllm_config, device)
         runner.vllm_config = vllm_config