[cuda][misc] remove error_on_invalid_device_count_status (#7069)

vllm/executor/multiproc_gpu_executor.py

@@ -17,7 +17,6 @@ from vllm.logger import init_logger
 from vllm.sequence import ExecuteModelRequest, SamplerOutput
 from vllm.triton_utils import maybe_set_triton_cache_manager
 from vllm.utils import (_run_task_with_lock, cuda_device_count_stateless,
-                        error_on_invalid_device_count_status,
                         get_distributed_init_method, get_open_port,
                         get_vllm_instance_id, make_async,
                         update_environment_variables)
@@ -79,8 +78,6 @@ class MultiprocessingGPUExecutor(DistributedGPUExecutor):
             f"please ensure that world_size ({world_size}) "
             f"is less than than max local gpu count ({cuda_device_count})")
 
-        error_on_invalid_device_count_status()
-
         # Multiprocessing-based executor does not support multi-node setting.
         # Since it only works for single node, we can use the loopback address
         # 127.0.0.1 for communication.

vllm/executor/ray_gpu_executor.py

@@ -10,10 +10,9 @@ from vllm.executor.distributed_gpu_executor import (  # yapf: disable
 from vllm.executor.ray_utils import RayWorkerWrapper, ray
 from vllm.logger import init_logger
 from vllm.sequence import ExecuteModelRequest, SamplerOutput
-from vllm.utils import (_run_task_with_lock,
-                        error_on_invalid_device_count_status,
-                        get_distributed_init_method, get_ip, get_open_port,
-                        get_vllm_instance_id, make_async)
+from vllm.utils import (_run_task_with_lock, get_distributed_init_method,
+                        get_ip, get_open_port, get_vllm_instance_id,
+                        make_async)
 
 if ray is not None:
     from ray.util.scheduling_strategies import PlacementGroupSchedulingStrategy
@@ -216,8 +215,6 @@ class RayGPUExecutor(DistributedGPUExecutor):
         distributed_init_method = get_distributed_init_method(
             driver_ip, get_open_port())
 
-        error_on_invalid_device_count_status()
-
         # Initialize the actual workers inside worker wrapper.
         init_worker_all_kwargs = [
             self._get_worker_kwargs(