BG：
之前在另外一台离线服务器（虚拟机），相同的conda环境，运行脚本无报错；conda环境迁移后，使用相同的环境和代码（python==3.9.0 easyocr==1.7.2）运行脚本。
日志发现最后一行报错如下：
[2025-12-16 05:35:53] [WARNING] Neither CUDA nor MPS are available - defaulting to CPU. Note: This module is much faster with a GPU. [W1216 05:36:01.540097613 NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
ps -ef | grep 查看py程序已经挂掉。
排查步骤：
1. 可能是easyocr有问题，但脚本执行没有问题；
(ocr) ubuntu@xcznmz-1112:/opt/lung_data$ python -<< 'EOF' >
import torch
> print("torch imported")
> x = torch.randn(2,3)
> y = torch.mm(x, x.T)
> print("torch mm ok")
> EOF
torch imported
torch mm ok
(ocr) ubuntu@xcznmz-1112:/opt/lung_data$ python -<< 'EOF' >
import easyocr
> reader = easyocr.Reader(['en'], model_storage_directory='./model', user_network_directory='./model', download_enabled =False, gpu =False)
> print("easyocr ok")
> EOF Using CPU. Note: This module is much faster with a GPU.
easyocr ok
2. github发现有相似报错，推荐降低torch版本
但是我是离线服务器，不好修改conda环境，不想动环境。
3. 排查CPU 指令集兼容性
lscpu | grep -E "Model name|Flags"
发现缺少CPU指令集 AVX，尝试脚本开头增加环境变量设置：
# @-*- coding: utf-8 -*-
import os
import sys
os.environ['NNPACK_DISABLE'] = '1'
os.environ['OMP_NUM_THREADS'] = '2' # 对于双核CPU
os.environ['MKL_NUM_THREADS'] = '2'
os.environ['OPENBLAS_NUM_THREADS'] = '2'
import resource resource.setrlimit(resource.RLIMIT_AS, (4 * 1024**3, 4 * 1024**3)) # 4GB虚拟内存限制
新的报错：
visit_date=2025-11-16 16:39:40, 错误=[enforce fail at alloc_cpu.cpp:117] err == 0. DefaultCPUAllocator: can't allocate memory: you tried to allocate 9437184 bytes. Error code 12 (Cannot allocate memory) Traceback (most recent call last): File "/opt/lung_data/lung_main.py", line 843, in process_patients result = ocr(record) File "/opt/lung_data/lung_main.py", line 285, in ocr reader_chi = easyocr.Reader(['ch_sim'], model_storage_directory='./model', user_network_directory='./model', download_enabled=False, gpu=False) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/easyocr/easyocr.py", line 214, in __init__ self.detector = self.initDetector(detector_path) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/easyocr/easyocr.py", line 271, in initDetector return self.get_detector(detector_path, File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/easyocr/detection.py", line 75, in get_detector net = CRAFT() File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/easyocr/craft.py", line 35, in __init__ self.basenet = vgg16_bn(pretrained, freeze) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/easyocr/model/modules.py", line 27, in __init__ vgg_pretrained_features = models.vgg16_bn( File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torchvision/models/_utils.py", line 142, in wrapper return fn(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torchvision/models/_utils.py", line 228, in inner_wrapper return builder(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torchvision/models/vgg.py", line 459, in vgg16_bn return _vgg("D", True, weights, progress, **kwargs) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torchvision/models/vgg.py", line 103, in _vgg model = VGG(make_layers(cfgs[cfg], batch_norm=batch_norm), **kwargs) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torchvision/models/vgg.py", line 81, in make_layers conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1) File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 521, in __init__ super().__init__( File "/home/ubuntu/anaconda3/envs/ocr/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 166, in __init__ torch.empty( RuntimeError: [enforce fail at alloc_cpu.cpp:117] err == 0. DefaultCPUAllocator: can't allocate memory: you tried to allocate 9437184 bytes. Error code 12 (Cannot allocate memory)
尝试修改代码，EasyOCR Reader 只创建一次，但是仍然出现上述两个报错。
4. 尝试将新服务器启用host cpu，重启后，运行脚本虽然还是报
Could not initialize NNPACK! Reason: Unsupported hardware.
但是py脚本并没有挂掉——解决！
将脚本开头的环境变量注释掉，仍然正常运行，未出现脚本挂掉的情况
import os import sys os.environ['NNPACK_DISABLE'] = '1' # os.environ['OMP_NUM_THREADS'] = '1' # 单线程最稳定 # os.environ['MKL_NUM_THREADS'] = '1' # os.environ['OPENBLAS_NUM_THREADS'] = '1' import resource # resource.setrlimit(resource.RLIMIT_AS, (8 * 1024**3, 8 * 1024**3)) # 8GB虚拟内存限制
)
