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 9, 01:15 UTC
[CI/Build] Investigate torchrun distributed tests hanging issue (#33650)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Feb 3, 2026 at 15:49 UTC
32e84fa1ff4d371d52042657a05d825c475cba3a
2 files changed
+6
tests/distributed/test_torchrun_example.py
+3
@@ -32,6 +32,9 @@ llm = LLM(
32
gpu_memory_utilization=random.uniform(0.7, 0.9),
33
swap_space=random.randint(1, 4),
34
seed=0,
35
+ # FIXME(Isotr0py): async scheduling causes deadlock
36
+ # on torchrun with PP, need to investigate further.
37
+ async_scheduling=False,
38
)
39
40
outputs = llm.generate(prompts, sampling_params)
tests/distributed/test_torchrun_example_moe.py
+3
@@ -39,6 +39,9 @@ llm = LLM(
39
gpu_memory_utilization=random.uniform(0.7, 0.9),
40
swap_space=random.randint(1, 4),
41
seed=0,
42
+ # FIXME(Isotr0py): async scheduling causes deadlock
43
+ # on torchrun with PP, need to investigate further.
44
+ async_scheduling=False,
45
)
46
47
outputs = llm.generate(prompts, sampling_params)