From 242d6018605f0299be2f4944cafeedf45ed98f50 Mon Sep 17 00:00:00 2001 From: Itay Alroy Date: Tue, 30 Dec 2025 15:56:53 +0200 Subject: [PATCH 1/2] nixl_ep: Unset UCX_NET_DEVICES Latest UCX auto-detects the correct NICs for each GPU Signed-off-by: Itay Alroy --- examples/device/ep/tests/elastic/elastic.py | 14 -------------- 1 file changed, 14 deletions(-) diff --git a/examples/device/ep/tests/elastic/elastic.py b/examples/device/ep/tests/elastic/elastic.py index 5a8daf9175..71bf0c5a4f 100644 --- a/examples/device/ep/tests/elastic/elastic.py +++ b/examples/device/ep/tests/elastic/elastic.py @@ -466,20 +466,6 @@ def worker(torch_rank: int, args: argparse.Namespace): torch.set_default_device("cuda") torch.cuda.set_device(0) - # Initialize UCX - pxb_nics = [ - "mlx5_0", - "mlx5_3", - "mlx5_4", - "mlx5_5", - "mlx5_6", - "mlx5_9", - "mlx5_10", - "mlx5_11", - ] - tcp_nics = ",ibp154s0,ibp192s0,ibp206s0,ibp220s0,ibp94s0" - os.environ["UCX_NET_DEVICES"] = f"cuda0-{pxb_nics[local_rank]}:1" + tcp_nics - # Initialize NIXL os.environ["NIXL_ETCD_ENDPOINTS"] = args.etcd_server From c29e5b16dcffbf9d6621163beb0e4ecf70cc9e5a Mon Sep 17 00:00:00 2001 From: Itay Alroy Date: Wed, 7 Jan 2026 13:10:36 +0200 Subject: [PATCH 2/2] Copyrights --- examples/device/ep/tests/elastic/elastic.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/device/ep/tests/elastic/elastic.py b/examples/device/ep/tests/elastic/elastic.py index 71bf0c5a4f..50360cac9f 100644 --- a/examples/device/ep/tests/elastic/elastic.py +++ b/examples/device/ep/tests/elastic/elastic.py @@ -1,5 +1,5 @@ # SPDX-FileCopyrightText: Copyright (c) 2025 DeepSeek -# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # This file incorporates material from the DeepSeek project, licensed under the MIT License. # The modifications made by NVIDIA are licensed under the Apache License, Version 2.0.