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
[XPU] Enable external_launcher to serve as an executor via torchrun (#21021)
Signed-off-by: chzhang <chaojun.zhang@intel.com>
Chaojun Zhang committed
Jul 22, 2025 at 12:47 UTC
25d585ab7bdf19009ab6685a33270401803b71a5
1 file changed
+3
-1
vllm/v1/worker/xpu_worker.py
+3
-1
@@ -7,6 +7,7 @@ import torch.distributed
7
8
import vllm.envs as envs
9
from vllm.config import VllmConfig
10
+from vllm.distributed import get_world_group
11
from vllm.logger import init_logger
12
from vllm.model_executor import set_random_seed
13
from vllm.platforms import current_platform
@@ -155,7 +156,8 @@ class XPUWorker(Worker):
156
current_platform.dist_backend)
157
158
# global all_reduce needed for overall oneccl warm up
158
- torch.distributed.all_reduce(torch.zeros(1).xpu())
159
+ torch.distributed.all_reduce(torch.zeros(1).xpu(),
160
+ group=get_world_group().device_group)
161
162
# Set random seed.
163
set_random_seed(self.model_config.seed)