device_manager.py

from functools import wraps
import torch
from huggingface_hub import HfApi
import os
import logging
import asyncio

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

class DeviceManager:
    _instance = None
    
    def __new__(cls):
        if cls._instance is None:
            cls._instance = super(DeviceManager, cls).__new__(cls)
            cls._instance._initialized = False
        return cls._instance
    
    def __init__(self):
        if self._initialized:
            return
            
        self._initialized = True
        self._current_device = None
        self._zero_gpu_available = None
        
    def check_zero_gpu_availability(self):
        try:
            if 'SPACE_ID' in os.environ:
                api = HfApi()
                space_info = api.get_space_runtime(os.environ['SPACE_ID'])
                if hasattr(space_info, 'hardware') and space_info.hardware.get('zerogpu', False):
                    self._zero_gpu_available = True
                    return True
        except Exception as e:
            logger.warning(f"Error checking ZeroGPU availability: {e}")
        
        self._zero_gpu_available = False
        return False
    
    def get_optimal_device(self):
        if self._current_device is None:
            if self.check_zero_gpu_availability():
                try:
                    self._current_device = torch.device('cuda')
                    logger.info("Using ZeroGPU")
                except Exception as e:
                    logger.warning(f"Failed to initialize ZeroGPU: {e}")
                    self._current_device = torch.device('cpu')
            else:
                self._current_device = torch.device('cpu')
                logger.info("Using CPU")
        return self._current_device

def device_handler(func):
    """簡化版的 device handler"""
    @wraps(func)
    async def wrapper(*args, **kwargs):
        device_mgr = DeviceManager()
        try:
            result = await func(*args, **kwargs)
            return result
        except RuntimeError as e:
            if "out of memory" in str(e) or "CUDA" in str(e):
                logger.warning("ZeroGPU unavailable, falling back to CPU")
                device_mgr._current_device = torch.device('cpu')
                return await func(*args, **kwargs)
            raise e
    return wrapper