A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 01:00 UTC
[Intel GPU] Ray Compiled Graph avoid NCCL for Intel GPU (#21338)
Signed-off-by: ratnampa <ratnam.parikh@intel.com>
Ratnam Parikh committed
Jul 21, 2025 at 21:48 UTC
90f1e55421f1b61394ba25abe34bf5abd82a71af
1 file changed
+2
-2
vllm/executor/ray_distributed_executor.py
+2
-2
@@ -67,8 +67,8 @@ class RayDistributedExecutor(DistributedExecutorBase):
67
os.environ["VLLM_USE_RAY_SPMD_WORKER"] = "1"
68
os.environ["VLLM_USE_RAY_COMPILED_DAG"] = "1"
69
70
- # For TPU, avoid compiling NVIDIA's NCCL
71
- if current_platform.is_tpu():
70
+ # For TPU or XPU, avoid compiling NVIDIA's NCCL
71
+ if current_platform.is_tpu() or current_platform.is_xpu():
72
os.environ["VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE"] = "shm"
73
74
# If the env var is set, it uses the Ray's compiled DAG API