PaddleX/zhch/utils/cuda_utils.py

import torch
from typing import List
def detect_available_gpus() -> List[int]:
    """检测可用的GPU"""
    try:
        gpu_count = torch.cuda.device_count()
        available_gpus = list(range(gpu_count))
        print(f"检测到 {gpu_count} 个可用GPU: {available_gpus}")
        return available_gpus
    except Exception as e:
        print(f"GPU检测失败: {e}")
        return []

def monitor_gpu_memory(gpu_ids: List[int] = [0, 1]):
    """监控GPU内存使用情况"""
    try:
        for gpu_id in gpu_ids:
            torch.cuda.set_device(gpu_id)
            total = torch.cuda.get_device_properties(gpu_id).total_memory / 1024**3
            allocated = torch.cuda.memory_allocated(gpu_id) / 1024**3
            reserved = torch.cuda.memory_reserved(gpu_id) / 1024**3
            print(f"GPU {gpu_id} - 显存: {total:.2f}GB, 已分配: {allocated:.2f}GB, 已预留: {reserved:.2f}GB")
    except Exception as e:
        print(f"GPU内存监控失败: {e}")