[BUG] torchrun subprocess received Signal 8 (SIGFPE)

[ZeXin-Wang]

Describe the bug

To Reproduce

conda env remove -n deepseek
conda create --name deepseek python=3.10
conda activate deepseek
pip install -r requirements.txt
python3 convert.py --hf-ckpt-path DeepSeek-V3 --save-path  DeepSeek-V3-Demo --n-experts 256 --model-parallel 8
# run on single machine with H20*8
torchrun --nnodes 1 --nproc-per-node 8 --node-rank 0 --master-addr 127.0.0.1 generate.py --ckpt-path DeepSeek-V3-Demo --config configs/config_671B.json --interactive --temperature 0.7 --max-new-tokens 200

[GeeeekExplorer]

There should be more detailed Python error messages before the torchrun error occurs. Also, which version of PyTorch and Triton do you use?