Method __init__

src/utils/gpu_monitor.py:9–49

初始化GPU显存监控器 Args: gpu_index: 要监控的GPU索引，默认0 interval: 监控间隔时间（秒），默认1秒 unit: 返回的单位，支持 'MB' 或 'GB'，默认'GB'

(self, gpu_index=0, interval=1.0, unit='GB')

class GPUMemoryMonitor:
    def __init__(self, gpu_index=0, interval=1.0, unit='GB'):
        """
        初始化GPU显存监控器
        
        Args:
            gpu_index: 要监控的GPU索引，默认0
            interval: 监控间隔时间（秒），默认1秒
            unit: 返回的单位，支持 'MB' 或 'GB'，默认'GB'
        """
        self.gpu_index = gpu_index
        self.interval = interval
        self.unit = unit.upper()
        
        # 验证单位参数
        if self.unit not in ['MB', 'GB']:
            raise ValueError("单位必须是 'MB' 或 'GB'")
        
        # 监控相关状态
        self.monitor_thread = None
        self._running = False
        self._lock = threading.Lock()
        self.peak_memory_usage = 0  # 峰值显存使用量
        self.start_time = None
        self.stop_time = None
        
        # 初始化NVML
        try:
            pynvml.nvmlInit()
            self.device_count = pynvml.nvmlDeviceGetCount()
            if self.gpu_index >= self.device_count:
                raise ValueError(f"GPU索引 {self.gpu_index} 超出范围，系统只有 {self.device_count} 个GPU")
            self.handle = pynvml.nvmlDeviceGetHandleByIndex(self.gpu_index)
            
            # 获取GPU名称
            self.gpu_name = pynvml.nvmlDeviceGetName(self.handle)
            
        except Exception as e:
            print(f"初始化NVML失败: {e}")
            if 'pynvml' in sys.modules:
                pynvml.nvmlShutdown()
            raise
    
    def _get_memory_info(self):
        """获取当前GPU显存信息"""