Split enable_mm (#7183)

Co-authored-by: liuruian <liuruian@MacBook-Pro.local>

tests/distributed/chunked_moe.py

@@ -92,6 +92,7 @@ class MockFDConfig:
     model_config = MockModelConfig()
     cache_config = MockCacheConfig()
     speculative_config = MockSpecaulativeConfig()
+    enable_mm_runtime = MockModelConfig.enable_mm
 
     def get_max_chunk_tokens(self, mm_max_tokens_per_item=None):
         return 8192
@@ -139,7 +140,7 @@ class TestChunkedMoE(unittest.TestCase):
         model_runner.model_config = mock_model_config
         model_runner.cache_config = mock_cache_config
         model_runner.attn_backends = [MockAttentionBackend()]
-        model_runner.enable_mm = True
+        model_runner.enable_mm = mock_fd_config.enable_mm_runtime
         model_runner.cudagraph_only_prefill = False
         model_runner.use_cudagraph = False
         model_runner.speculative_decoding = False

tests/entrypoints/test_engine_client.py

@@ -102,6 +102,7 @@ def create_mock_fd_config(
     mock_config.structured_outputs_config = Mock()
     mock_config.structured_outputs_config.reasoning_parser = None
     mock_config.tool_parser = None
+    mock_config.enable_mm_runtime = enable_mm
 
     return mock_config
 
@@ -181,6 +182,7 @@ class TestEngineClient(unittest.IsolatedAsyncioTestCase):
         mock_config.structured_outputs_config = Mock()
         mock_config.structured_outputs_config.reasoning_parser = None
         mock_config.node_rank = 0
+        mock_config.enable_mm_runtime = mock_model_config.enable_mm
 
         # Create mocks for all the external dependencies
         mock_input_processor = Mock()
@@ -363,6 +365,7 @@ class TestEngineClientValidParameters(unittest.TestCase):
         mock_config.structured_outputs_config = MagicMock()  # Add this
         mock_config.structured_outputs_config.reasoning_parser = None
         mock_config.tool_parser = None  # Add this attribute
+        mock_config.enable_mm_runtime = mock_model_config.enable_mm
 
         # Mock IPCSignal to avoid file system dependencies
         with patch("fastdeploy.entrypoints.engine_client.IPCSignal") as mock_ipcsignal:
@@ -655,6 +658,7 @@ class TestEngineClientValidParameters(unittest.TestCase):
             mock_config.structured_outputs_config = Mock()
             mock_config.structured_outputs_config.reasoning_parser = None
             mock_config.tool_parser = None
+            mock_config.enable_mm_runtime = mock_config.model_config.enable_mm
 
             client = EngineClient(
                 pid=5678,
@@ -1078,6 +1082,7 @@ class TestEngineClientValidParameters(unittest.TestCase):
 
         mock_config = Mock()
         mock_config.model_config = mock_model_config
+        mock_config.enable_mm_runtime = mock_model_config.enable_mm
         mock_config.eplb_config = Mock()
         mock_config.eplb_config.enable_eplb = False
 
@@ -1131,6 +1136,7 @@ class TestEngineClientValidParameters(unittest.TestCase):
 
         mock_config = Mock()
         mock_config.model_config = mock_model_config
+        mock_config.enable_mm_runtime = mock_model_config.enable_mm
         mock_config.eplb_config = Mock()
         mock_config.eplb_config.enable_eplb = False
 
@@ -1408,6 +1414,7 @@ class TestEngineClientValidParameters(unittest.TestCase):
 
         mock_config = Mock()
         mock_config.model_config = mock_model_config
+        mock_config.enable_mm_runtime = mock_model_config.enable_mm
         mock_config.eplb_config = Mock()
         mock_config.eplb_config.enable_eplb = False
 

tests/layers/test_kv_cache_int8_dynamic_quant_backend.py

@@ -92,6 +92,7 @@ class DummyFDConfig:
                 "max_model_len": 2048,
                 "head_dim": 128,
                 "num_hidden_layers": 2,
+                "enable_mm": False,
                 "causal": True,
                 "start_layer_index": 0,
                 "rope_3d": False,
@@ -124,6 +125,8 @@ class DummyFDConfig:
                 "model_type": "main",
             },
         )()
+        self.enable_mm_runtime = self.model_config.enable_mm
+        self.enable_rope_3d_runtime = self.model_config.enable_mm
 
 
 class DummyLayer:

tests/scheduler/test_chunked_prefill_determinism.py

@@ -78,6 +78,7 @@ class StubConfig:
         self.cache_config = CacheConfig()
         self.parallel_config = ParallelConfig()
         self.speculative_config = SpeculativeConfig()
+        self.enable_mm_runtime = self.model_config.enable_mm
 
 
 # ---------------------------------------------------------------------------
@@ -168,6 +169,7 @@ class TestChunkedPrefillDeterminism(unittest.TestCase):
     def _create_mm_resource_manager(self):
         config = StubConfig()
         config.model_config.enable_mm = True
+        config.enable_mm_runtime = config.model_config.enable_mm
         return self._create_resource_manager(config)
 
     # ==================== 1. Deterministic disabled ====================

tests/worker/test_gpu_prompt_logprobs.py

@@ -64,6 +64,7 @@ class MockConfig:
     scheduler_config = SchedulerConfig()
     cache_config = CacheConfig()
     parallel_config = ParallelConfig()
+    enable_mm_runtime = model_config.enable_mm
 
     def get_max_chunk_tokens(self, mm_max_tokens_per_item=None):
         return 8192

tests/worker/test_reorder_split_prefill_and_decode.py

@@ -83,6 +83,7 @@ def create_mock_config():
     fd_config.parallel_config = parallel_config
     fd_config.structured_outputs_config = structured_outputs_config
     fd_config.pad_to = 8
+    fd_config.enable_mm_runtime = model_config.enable_mm
 
     def get_max_chunk_tokens(mm_max_tokens_per_item=None):
         return 100