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 13, 05:45 UTC
use default CCL_ZE_IPC_EXCHANGE (#28700)
Signed-off-by: Yan Ma <yan.ma@intel.com>
Yan Ma committed
Nov 14, 2025 at 16:55 UTC
529cea343da8662f135a69d9c3157f388f5eb64a
1 file changed
-2
vllm/v1/worker/xpu_worker.py
-2
@@ -159,12 +159,10 @@ class XPUWorker(Worker):
159
else:
160
raise RuntimeError(f"Not support device type: {self.device_config.device}")
161
162
- ENV_CCL_ZE_IPC_EXCHANGE = os.getenv("CCL_ZE_IPC_EXCHANGE", "pidfd")
162
ENV_CCL_ATL_TRANSPORT = os.getenv("CCL_ATL_TRANSPORT", "ofi")
163
ENV_LOCAL_WORLD_SIZE = os.getenv(
164
"LOCAL_WORLD_SIZE", str(self.parallel_config.world_size)
165
)
167
- os.environ["CCL_ZE_IPC_EXCHANGE"] = ENV_CCL_ZE_IPC_EXCHANGE
166
os.environ["CCL_ATL_TRANSPORT"] = ENV_CCL_ATL_TRANSPORT
167
os.environ["LOCAL_WORLD_SIZE"] = ENV_LOCAL_WORLD_SIZE
168
os.environ["LOCAL_RANK"] = str(self.local_rank)