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 8, 00:00 UTC
[AMD][CI] Fix test_custom_allreduce for A100 testgroup (#34735)
Signed-off-by: Ryan Rock <ryan.rock@amd.com>
Ryan Rock committed
Feb 20, 2026 at 15:33 UTC
0632ed8778cab44de6152eb873d09fa40c241962
1 file changed
+2
tests/distributed/test_custom_all_reduce.py
+2
@@ -33,6 +33,7 @@ def graph_allreduce(
33
):
34
with monkeypatch.context() as m:
35
m.delenv("CUDA_VISIBLE_DEVICES", raising=False)
36
+ m.delenv("HIP_VISIBLE_DEVICES", raising=False)
37
device = torch.device(f"cuda:{rank}")
38
torch.cuda.set_device(device)
39
init_test_distributed_environment(tp_size, pp_size, rank, distributed_init_port)
@@ -92,6 +93,7 @@ def eager_allreduce(
93
):
94
with monkeypatch.context() as m:
95
m.delenv("CUDA_VISIBLE_DEVICES", raising=False)
96
+ m.delenv("HIP_VISIBLE_DEVICES", raising=False)
97
device = torch.device(f"cuda:{rank}")
98
torch.cuda.set_device(device)
99
init_test_distributed_environment(tp_size, pp_size, rank, distributed_init_port)