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, 02:15 UTC
[Intel GPU] support ray as distributed executor backend for XPU. (#20659)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Kunshang Ji committed
Jul 9, 2025 at 15:36 UTC
b6e7e3d58f57aee30a55b3160645ddb2f029d3c8
2 files changed
+3
-1
.buildkite/scripts/hardware_ci/run-xpu-test.sh
+2
@@ -27,6 +27,8 @@ docker run \
27
"${image_name}" \
28
sh -c '
29
VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
30
+ VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend ray
31
+ VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend mp
32
cd tests
33
pytest -v -s v1/core
34
'
vllm/executor/ray_distributed_executor.py
+1
-1
@@ -62,7 +62,7 @@ class RayDistributedExecutor(DistributedExecutorBase):
62
63
def _init_executor(self) -> None:
64
self.forward_dag: Optional[ray.dag.CompiledDAG] = None
65
- if envs.VLLM_USE_V1 and not current_platform.is_xpu():
65
+ if envs.VLLM_USE_V1:
66
# V1 uses SPMD worker and compiled DAG
67
os.environ["VLLM_USE_RAY_SPMD_WORKER"] = "1"
68
os.environ["VLLM_USE_RAY_COMPILED_DAG"] = "1"