[Bug]: external_launcher下无法感知data_parallel_size 导致cpu_group计算异常 assert self.cpu_group is not None

[tardis-key]

Your current environment

The output of `python collect_env.py`

PyTorch version: 2.5.1+cpu
Is debug build: False

OS: Debian GNU/Linux 10 (buster) (x86_64)
GCC version: (GCC) 9.3.0
Clang version: Could not collect
CMake version: version 4.0.1
Libc version: glibc-2.28

Python version: 3.10.14 (main, Jul 3 2024, 11:05:37) [GCC 8.3.0] (64-bit runtime)
Python platform: Linux-5.10.135.bsk.6-amd64-x86_64-with-glibc2.28

Versions of relevant libraries:
[pip3] gpytorch==1.14
[pip3] numpy==1.26.4
[pip3] pyzmq==26.4.0
[pip3] torch==2.5.1+cpu
[pip3] torch-npu==2.5.1.dev20250320
[pip3] torchaudio==2.5.1+cpu
[pip3] torchdata==0.11.0
[pip3] torchvision==0.20.1+cpu
[pip3] transformers==4.52.4
[conda] Could not collect
vLLM Version: 0.9.0
vLLM Ascend Version: 0.9.0rc2.dev0+g706de02.d20250610 (git sha: 706de02, date: 20250610)

CANN:
package_name=Ascend-cann-toolkit
version=8.1.RC1
innerversion=V100R001C21SPC001B203
compatible_version=[V100R001C15],[V100R001C18],[V100R001C19],[V100R001C20],[V100R001C21]
arch=x86_64
os=linux
path=/usr/local/Ascend/ascend-toolkit/8.1.RC1/x86_64-linux

🐛 Describe the bug

`self.inference_engine = LLM( model=model_path, enable_sleep_mode=True, tensor_parallel_size=tensor_parallel_size, distributed_executor_backend="external_launcher", # 0.9.0推新的dp方案，ascend适配过程没到位,在el下没有感知dp data_parallel_size=8, # 手动修改，需要输入dp size dtype=config.dtype, enforce_eager=config.enforce_eager, gpu_memory_utilization=config.gpu_memory_utilization, disable_custom_all_reduce=True, disable_mm_preprocessor_cache=False, skip_tokenizer_init=False, max_model_len=max_model_len, load_format=load_format, disable_log_stats=config.disable_log_stats, max_num_batched_tokens=max_num_batched_tokens, enable_chunked_prefill=config.enable_chunked_prefill, enable_prefix_caching=True, trust_remote_code=trust_remote_code, seed=config.get("seed", 0), **lora_kwargs, **engine_kwargs, )`

[HIT-leaderone]

怎么修改啊

[MengqingCao]

The temporary solution is to manually pass in data_parallel_size when creating the LLM object in the RL framework. We will discuss the long-term solution later.

[shuguang99]

然而，设置了data_parallel_size=8 tensor_parallel_size=1(8卡单节点)，程序会卡住，是我的设置问题吗？

[leo-pony]

然而，设置了data_parallel_size=8 tensor_parallel_size=1(8卡单节点)，程序会卡住，是我的设置问题吗？

@shuguang99
Plz give out the using vllm, vllm-ascend, and verl version, and plz give out the hang stack information.

[Yikun]

This issue will be resolved by:
v0.9.1-dev ---> v0.9.1rc2 #1831
main ---> v0.10.1rc1 #1681