Mac M1设备Conda环境安装TensorFlow 2.5时NumPy导入报错兼容问题

操作复现步骤 #

• 安装适配arm64架构的miniforge3版本
• 执行命令创建并激活Conda环境：

  conda create --name tf25
  conda activate tf25
  

• 完成苹果官方TensorFlow插件配置流程
• 运行测试代码：

  import numpy
  

报错信息 #

Traceback (most recent call last): File "/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/init.py", line 22, in
from . import multiarray File "/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/multiarray.py", line 12, in
from . import overrides File "/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/overrides.py", line 7, in
from numpy.core._multiarray_umath import ( ImportError: dlopen(/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/_multiarray_umath.cpython-39-darwin.so, 2): Library not loaded: @rpath/libcblas.3.dylib Referenced from: /Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/_multiarray_umath.cpython-39-darwin.so
Reason: image not found

处理上述异常时触发二次报错：
Traceback (most recent call last): File "", line 1, in
File "/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/init.py", line 140, in
from . import core File "/Users/antoniosquicciarini/miniforge3/envs/tf25/lib/python3.9/site-packages/numpy/core/init.py", line 48, in
raise ImportError(msg) ImportError:

对应环境信息：

• Python版本：3.9，路径为/Users/antoniosquicciarini/miniforge3/envs/tf25/bin/python
• NumPy版本：1.19.5
  根因：libcblas.3.dylib依赖库缺失，属于M1平台下TensorFlow 2.5与默认安装的NumPy版本兼容性问题。

已验证无效的解决方案 #

• 直接通过pip安装、重装NumPy
• 降级Python到3.8版本

可行修复步骤 #

激活tf25环境后依次执行以下操作：

1. 彻底卸载当前环境内的所有NumPy版本：

pip uninstall -y numpy
conda uninstall -y numpy

2. 安装conda-forge源的OpenBLAS依赖，补全缺失的cblas库：

conda install -c conda-forge openblas

3. 安装适配M1架构、兼容TensorFlow 2.5的指定NumPy版本：

conda install -c conda-forge numpy==1.21.2

4. 运行import numpy测试，无报错即为修复完成。如果仍存在兼容性问题，可重装对应版本的TensorFlow依赖：

conda install -c apple tensorflow-deps==2.5.0
pip install tensorflow-macos==2.5.0
pip install tensorflow-metal==0.1.2

内容的提问来源于stack exchange，提问作者Antonio