ModelScope中，修改了代码还是报错怎么解决？

"单卡内存不足，多卡的话，报错RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method。ModelScope中，修改了代码还是报错怎么解决？ import os
import subprocess
import multiprocessing as mp

if name == 'main':

# 设置多处理的启动方法为'spawn'
mp.set_start_method('spawn')

# 设置环境变量
os.environ['CUDA_VISIBLE_DEVICES'] = '2,3,4,5'
os.environ['RAY_memory_monitor_refresh_ms'] = '0'

# 运行swift deploy命令
subprocess.run([
    'swift', 'app-ui',
    '--model_type', 'qwen2-7b',
    '--infer_backend', 'vllm',
    '--tensor_parallel_size', '4',
    '--dtype', 'fp16'
]) 还是报错RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method"