fix mla ut

Signed-off-by: wangli <wangli858794774@gmail.com>

Author: Potabk

tests/ut/torchair/test_torchair_mla.py

@@ -180,18 +180,19 @@ def test_ascend_mla_metadata_default(self):
 class TestAscendMLATorchairMetadataBuilder(TestBase):
 
     def test_ascend_mla_metadata_builder_default(self):
-        mock_vllm_config = MagicMock()
-        mock_vllm_config.model_config.max_model_len = 1024
-        mock_vllm_config.model_config.get_head_size.return_value = 64
-        mock_vllm_config.model_config.dtype = torch.float16
-        mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.get_head_size = lambda: 8
-        mock_vllm_config.scheduler_config.max_num_seqs = 4
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
-        mock_device = torch.device('cpu')
+        mock_model_config = MagicMock()
+        mock_model_config.max_model_len = 1024
+        mock_model_config.get_head_size.return_value = 64
+        mock_model_config.dtype = torch.float16
 
+        mock_vllm_config = MagicMock()
+        mock_vllm_config.model_config = mock_model_config
+        mock_vllm_config.cache_config = MagicMock(block_size=16)
+        mock_vllm_config.scheduler_config = MagicMock(
+            max_num_seqs=4, enable_chunked_prefill=False)
         mock_vllm_config.speculative_config = None
 
+        mock_device = torch.device('cpu')
         ascend_config = MagicMock()
         ascend_config.torchair_graph_config = MagicMock()
         ascend_config.torchair_graph_config.enabled = True
@@ -205,23 +206,25 @@ def test_ascend_mla_metadata_builder_default(self):
                              mock_vllm_config.cache_config.block_size)
             self.assertEqual(
                 builder.chunked_prefill_enabled,
-                mock_vllm_config.scheduler_config.chunked_prefill_enabled)
+                mock_vllm_config.scheduler_config.enable_chunked_prefill)
             self.assertEqual(builder.torchair_graph_enabled, True)
 
     @patch("vllm_ascend.torchair.torchair_mla.get_ascend_config")
     def test_reorder_batch_with_torchair_graph(self, ascend_config):
-        mock_vllm_config = MagicMock()
-        mock_vllm_config.model_config.max_model_len = 1024
-        mock_vllm_config.get_head_size = lambda: 8
-        mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.scheduler_config.max_num_seqs = 4
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
-        mock_device = torch.device('cpu')
-        ascend_config.torchair_graph_config = MagicMock()
-        ascend_config.torchair_graph_config.enabled = True
+        mock_model_config = MagicMock()
+        mock_model_config.max_model_len = 1024
+        mock_model_config.get_head_size.return_value = 64
+        mock_model_config.dtype = torch.float16
 
+        mock_vllm_config = MagicMock()
+        mock_vllm_config.model_config = mock_model_config
+        mock_vllm_config.cache_config = MagicMock(block_size=16)
+        mock_vllm_config.scheduler_config = MagicMock(
+            max_num_seqs=4, enable_chunked_prefill=False)
         mock_vllm_config.speculative_config = None
 
+        mock_device = torch.device('cpu')
+
         builder = AscendMLATorchairMetadataBuilder(None, None,
                                                    mock_vllm_config,
                                                    mock_device)
@@ -255,7 +258,7 @@ def test_reorder_batch_without_torchair_graph(self):
         mock_vllm_config.get_head_size = lambda: 8
         mock_vllm_config.cache_config.block_size = 16
         mock_vllm_config.scheduler_config.max_num_seqs = 4
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_device = torch.device('cpu')
 
         mock_vllm_config.speculative_config = None
@@ -293,7 +296,7 @@ def test_get_graph_runner_block_tables_normal(self, mock_ascend_config):
         mock_vllm_config = MagicMock()
         mock_vllm_config.model_config.max_model_len = 1024
         mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_device = torch.device('cpu')
 
         mock_vllm_config.speculative_config = None
@@ -316,7 +319,7 @@ def test_get_graph_runner_block_tables_truncated(self, mock_ascend_config):
         mock_vllm_config = MagicMock()
         mock_vllm_config.model_config.max_model_len = 64
         mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_device = torch.device('cpu')
 
         mock_vllm_config.speculative_config = None
@@ -342,7 +345,7 @@ def test_get_graph_runner_block_tables_from_numpy(self,
         mock_vllm_config.cache_config.block_size = 16
         mock_vllm_config.get_head_size = lambda: 28
         mock_vllm_config.dtype = torch.bfloat16
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_device = torch.device('cpu')
 
         mock_vllm_config.speculative_config = None
@@ -368,7 +371,7 @@ def test_build_dummy(self, mock_ascend_config):
         mock_vllm_config = MagicMock()
         mock_vllm_config.model_config.max_model_len = 1024
         mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_vllm_config.get_head_size.return_value = 64
         mock_vllm_config.model_config.dtype = torch.float16
         mock_device = torch.device('cpu')
@@ -435,7 +438,7 @@ def test_build_decode(self, mock_ascend_config):
         mock_vllm_config = MagicMock()
         mock_vllm_config.model_config.max_model_len = 1024
         mock_vllm_config.cache_config.block_size = 16
-        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_vllm_config.scheduler_config.enable_chunked_prefill = False
         mock_vllm_config.get_head_size.return_value = 64
         mock_vllm_config.model_config.dtype = torch.float16
         mock_device = torch.device('cpu')

vllm_ascend/utils.py

@@ -471,9 +471,11 @@ def update_aclgraph_sizes(vllm_config: VllmConfig) -> None:
 
     # Calculate parallel configuration factor
     if not vllm_config.model_config:
-        logger.warning("Got empty model config, This occurs in scenarios \
-            where an empty config needs to be initialized, eg: unit tests, \
-                where updates are skipped.")
+        logger.warning(
+            "Got empty model config. This typically occurs when an empty vllm_config is "
+            "initialized (e.g., in unit tests), where config updates are intentionally skipped."
+        )
+
         return
     hf_config = vllm_config.model_config.hf_config
     if hasattr(hf_config, 'num_hidden_layers'):