[Bugfix] No num_gpus for ROCm and XPU when connecting to a ray cluster

I don't find any reason for making a special case for ROCm and XPU,
as they should make no difference with CUDA or something else in this
case.

Error log:

  File "vllm/vllm/engine/llm_engine.py", line 528, in _get_executor_cls
    initialize_ray_cluster(engine_config.parallel_config)
  File "vllm/vllm/executor/ray_utils.py", line 230, in initialize_ray_cluster
    ray.init(address=ray_address,
  File "venv/lib/python3.11/site-packages/ray/_private/client_mode_hook.py",
   line 103, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "venv/lib/python3.11/site-packages/ray/_private/worker.py",
   line 1689, in init
    raise ValueError(
ValueError: When connecting to an existing cluster, num_cpus and num_gpus must not be provided.

Signed-off-by: Hollow Man <[email protected]>

vllm/executor/ray_utils.py

@@ -10,7 +10,7 @@
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
 from vllm.sequence import ExecuteModelRequest, IntermediateTensors
-from vllm.utils import get_ip, is_hip, is_xpu
+from vllm.utils import get_ip
 from vllm.worker.worker_base import WorkerWrapperBase
 
 logger = init_logger(__name__)
@@ -226,12 +226,7 @@ def initialize_ray_cluster(
     assert_ray_available()
 
     # Connect to a ray cluster.
-    if is_hip() or is_xpu():
-        ray.init(address=ray_address,
-                 ignore_reinit_error=True,
-                 num_gpus=parallel_config.world_size)
-    else:
-        ray.init(address=ray_address, ignore_reinit_error=True)
+    ray.init(address=ray_address, ignore_reinit_error=True)
 
     if parallel_config.placement_group:
         # Placement group is already set.