[ROCm][CI] Give the MoE parallel-launch watchdog room for AITER JIT builds (#62161)

Signed-off-by: Andreas Karatzas <AndreasKaratzas@users.noreply.github.com>
Co-authored-by: Andreas Karatzas <AndreasKaratzas@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Andreas Karatzas
2026-10-17 21:49:27 -05:00
committed by GitHub
co-authored by Andreas Karatzas Claude
parent b2456c622b
commit da9134bc89
@@ -18,13 +18,16 @@ from vllm.distributed import (
init_distributed_environment,
initialize_model_parallel,
)
from vllm.platforms import current_platform
from vllm.utils.network_utils import get_open_port
## Parallel Processes Utils
P = ParamSpec("P")
LAUNCH_TIMEOUT_S = 900
# On ROCm the first AITER CK MoE launch JIT-compiles two stage modules that the
# AITER wheel does not prebuild, ~10 min each (ROCm/aiter#5766).
LAUNCH_TIMEOUT_S = 1800 if current_platform.is_rocm() else 900
@dataclasses.dataclass