test/npu/helloworld.py

import torch
import moxing as mox
import os
import argparse
import torch_npu  # 确保安装了 torch_npu

# 创建参数解析器
parser = argparse.ArgumentParser(description='Training script with output path option')
parser.add_argument('--dataset_input', default=None, type=str, help='dataset_input, where the dataset is stored.')
parser.add_argument('--output', type=str, default='', help='Output path for saving the result')
args = parser.parse_args()

# 检查 NPU 是否可用
try:
    import torch_npu

    if torch_npu.npu.is_available():
        device = torch.device('npu')
        print("Using NPU for training.")
    else:
        device = torch.device('cpu')
        print("NPU is not available, using CPU for training.")
except ImportError:
    device = torch.device('cpu')
    print("torch_npu module not found, using CPU for training.")

# 模拟一个简单的训练过程
# 这里其实不涉及真正的模型训练，只是作为示例流程展示
# 输出 Hello World
result = "Hello World"
print(result)

# 将结果保存到本地文件
local_result_path = 'hellworld.pth'
with open(local_result_path, 'w') as f:
    f.write(result)

# 根据参数确定 OBS 路径
obs_result_path = args.output if args.output else 'obs://nudt-cloudream2/cds/trainResult/model/hellworld.pth'
# 如果传入的路径不是以斜杠结尾，则添加斜杠
if obs_result_path and not obs_result_path.endswith('/'):
    obs_result_path += '/'
obs_result_path += 'hellworld.pth'

try:
    mox.file.copy(local_result_path, obs_result_path)
    print(f"Result has been successfully saved to {obs_result_path}")
except Exception as e:
    print(f"Failed to save result to OBS: {e}")