import clip时Cannot re-initialize CUDA in forked subprocess

问题

pytorch在项目里 pip install clip之后 import clip报错（infer时）：
“Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the ‘spawn’ start method”

解决

查阅位置跟multi process有关

for i in range(thread_num):
        p = mp.Process(target=sub_processor, args=(lock, i, args, data, 
                                                   save_path_prefix, save_visualize_path_prefix, 
                                                   img_folder, sub_video_list))

网上解决办法：

torch.multiprocessing.set_start_method(‘spawn’)

修改之后报错：

TypeError: cannot pickle ‘_thread.lock’ object

==============================================================================
逐行排查代码后发现与上述mp无关，把CLIP的load函数的参数从：

def load(name: str, device: Union[str, torch.device] = “cuda” if torch.cuda.is_available() else “cpu”, …

修改为（直接默认为cuda了 因为我暂时没有cpu的需求）

def load(name: str, device: Union[str, torch.device] = “cuda”, …