导入OCPCalculator时出现OSError错误求助

卸载原有PyTorch版本并安装适配任务的版本后，导入OCPCalculator时仍触发OSError，具体代码及错误信息如下：

代码 #

from fairchem.core.common.relaxation.ase_utils import OCPCalculator
import ase.io
from ase.optimize import BFGS

错误信息 #

OSError   Traceback (most recent call last)/var/folders/n7/96_5cstj2sn6g5ll0b979w380000gn/T/ipykernel_49125/1241704498.py in <module>
----> 1 from fairchem.core.common.relaxation.ase_utils import OCPCalculator
  2 import ase.io
  3 from ase.optimize import BFGS
~/opt/anaconda3/lib/python3.9/site-packages/fairchem/core/common/relaxation/ase_utils.py in <module>
 24 
 25 from fairchem.core.common.registry import registry
---> 26 from fairchem.core.common.utils import (
 27     load_config,
 28     setup_imports,

~/opt/anaconda3/lib/python3.9/site-packages/fairchem/core/common/utils.py in <module>
 30 import torch
 31 import torch.nn as nn
---> 32 import torch_geometric
~/opt/anaconda3/lib/python3.9/sitepackages/torch_geometric/__init__.py in <module>
 11 import torch_geometric.loader
 12 import torch_geometric.transforms
---> 13 import torch_geometric.datasets
...
384             self._handle = handle

OSError: 
dlopen(/Users/roshnidantuluri/opt/anaconda3/lib/python3.9/sitepackages/torch_cluster/_version_cpu.so, 0x0006): 
Symbol not found: __ZN3c1017RegisterOperatorsD1Ev
Referenced from: <6A95460A-9602-33F2-92B2-6A9C90B59457>/Users/roshnidantuluri/opt/anaconda3/lib/python3.9/site-packages/torch_cluster/_version_cpu.so
Expected in:     <709C1DF5-D253-3C66-87E2-C99FD3A259DF>/Users/roshnidantuluri/opt/anaconda3/lib/python3.9/site-packages/torch/lib/libtorch_cpu.dylib

• 核心原因：torch_cluster编译时依赖的PyTorch版本与当前安装的PyTorch版本不兼容，导致符号缺失。
• 步骤1：完全卸载相关依赖
  执行以下命令清理所有PyTorch及PyG生态库：

  pip uninstall -y torch torchvision torchaudio torch_geometric torch_cluster torch_scatter torch_sparse torch_spline_conv
  

• 步骤2：重新安装匹配版本的依赖
  针对macOS + Python3.9环境，按顺序执行：
  1. 安装适配的PyTorch稳定版：

     pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2
     

  2. 安装与PyTorch版本严格匹配的PyG依赖库：选择对应PyTorch版本的预编译包进行安装，执行：

     pip install torch_cluster==2.1.1 torch_scatter==2.1.1 torch_sparse==0.6.17 torch_spline_conv==1.2.2
     

  3. 安装对应版本的torch_geometric：

     pip install torch_geometric==2.3.1
     

• 步骤3：验证安装
  重新运行导入代码，检查是否仍有错误。
• 备选方案：创建干净的conda环境
  若原有环境存在依赖冲突，建议新建环境重新安装：

  conda create -n ocp_env python=3.9
  conda activate ocp_env
  # 重复步骤2的安装命令
  

内容的提问来源于stack exchange，提问作者Roshini Dantuluri