qwen3-vl-30b的多机多卡megatron-swift grpo训练结束时报错异常退出

所有steps都训练完成之后，异常退出了。这个是报错：traceback : Signal 6 (SIGABRT) received by PID 592
error_file:
exitcode : -6 (pid: 592)
rank : 19 (local_rank: 3)
host : cce-ig-jhi6i62h-s5pp6kok-k049m9py-03
time : 2025-12-02_13:14:06
[0]:
Root Cause (first observed failure):

Failures:

/mnt/volumes/ss-sai-bd-ga/liupengyi/ms-swift/swift/cli/_megatron/rlhf.py FAILED

torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
raise ChildFailedError(
File "/usr/local/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 277, in launch_agent
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
return launch_agent(self._config, self._entrypoint, list(args))
File "/usr/local/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 143, in call
elastic_launch(
File "/usr/local/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in run
run(args)
File "/usr/local/lib/python3.11/site-packages/torch/distributed/run.py", line 901, in main
^^^^^^^^^^^^^^^^^^
return f(*args, **kwargs)
File "/usr/local/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/init.py", line 357, in wrapper
main()
File "/us

Start with swift/cli/_megatron/rlhf.py and the PyTorch distributed launcher context around the reported rank 19 SIGABRT after all training steps complete. Reproduce the multi-node, multi-GPU Qwen3-VL GRPO run if the missing configuration can be obtained; done means the completed training exits normally without ChildFailedError and the underlying failure is identified.