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
[BugFix][Intel GPU] Use refactored API for dist_backend in V1 worker (#20596)
Signed-off-by: ratnampa <ratnam.parikh@intel.com>
Ratnam Parikh committed
Jul 8, 2025 at 19:44 UTC
c40784c7947acc247e857643d1046335b6d547cd
1 file changed
+2
-2
vllm/v1/worker/xpu_worker.py
+2
-2
@@ -148,11 +148,11 @@ class XPUWorker(Worker):
148
os.environ["CCL_ATL_TRANSPORT"] = ENV_CCL_ATL_TRANSPORT
149
os.environ["LOCAL_WORLD_SIZE"] = ENV_LOCAL_WORLD_SIZE
150
os.environ["LOCAL_RANK"] = str(self.local_rank)
151
- dist_backend = "ccl"
151
152
init_worker_distributed_environment(self.vllm_config, self.rank,
153
self.distributed_init_method,
155
- self.local_rank, dist_backend)
154
+ self.local_rank,
155
+ current_platform.dist_backend)
156
157
# global all_reduce needed for overall oneccl warm up
158
torch.distributed.all_reduce(torch.zeros(1).xpu())