[VLM] Refactor MultiModalConfig initialization and profiling (#7530)

vllm/executor/ray_xpu_executor.py

@@ -7,9 +7,8 @@ from typing import (TYPE_CHECKING, Any, Awaitable, Dict, List, Optional, Set,
 
 import vllm.envs as envs
 from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
-                         ModelConfig, MultiModalConfig, ParallelConfig,
-                         PromptAdapterConfig, SchedulerConfig,
-                         SpeculativeConfig)
+                         ModelConfig, ParallelConfig, PromptAdapterConfig,
+                         SchedulerConfig, SpeculativeConfig)
 from vllm.executor.distributed_gpu_executor import (  # yapf: disable
     DistributedGPUExecutor, DistributedGPUExecutorAsync)
 from vllm.executor.ray_utils import RayWorkerWrapper, ray
@@ -46,7 +45,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
         device_config: DeviceConfig,
         load_config: LoadConfig,
         lora_config: Optional[LoRAConfig],
-        multimodal_config: Optional[MultiModalConfig],
         prompt_adapter_config: Optional[PromptAdapterConfig],
         speculative_config: Optional[SpeculativeConfig],
     ) -> None:
@@ -61,7 +59,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
         self.parallel_config = parallel_config
         self.scheduler_config = scheduler_config
         self.device_config = device_config
-        self.multimodal_config = multimodal_config
         self.prompt_adapter_config = prompt_adapter_config
 
         placement_group = self.parallel_config.placement_group
@@ -203,7 +200,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
                     rank=rank,
                     distributed_init_method=distributed_init_method,
                     lora_config=self.lora_config,
-                    multimodal_config=self.multimodal_config,
                     is_driver_worker=rank == 0,
                 ))
         self._run_workers("init_worker", all_kwargs=init_worker_all_kwargs)