TensorFlow CPU版导入失败（DLL load failed），重装后仍异常求助

Hey there, this kind of leftover clutter from a failed TensorFlow GPU install is really common—let’s walk through a thorough cleanup and reinstall to get your CPU version working again.

Step 1: Fully Purge TensorFlow and Related Residues #

First, we need to make sure there’s no leftover GPU or broken CPU files hanging around:

• Uninstall all TensorFlow variants via pip:

  pip uninstall -y tensorflow tensorflow-gpu tensorflow-estimator tensorflow-intel keras keras-core keras-preprocessing
  

• Manually delete leftover folders in your Python environment’s Lib\site-packages directory. Look for folders named tensorflow, keras, keras_core, and any related subfolders—delete them all.
• Clean up registry entries (Windows only):
  1. Open regedit via the Start Menu.
  2. Navigate to HKEY_CURRENT_USER\Software\Python\PythonCore\[Your Python Version]\Packages and delete any TensorFlow/Keras-related keys.
  3. Repeat the check in HKEY_LOCAL_MACHINE\Software\Python\PythonCore\[Your Python Version]\Packages (if present) and remove relevant entries. Be careful not to delete unrelated keys!

Step 2: Fix Dependencies and System Requirements #

Most DLL failures stem from missing system components or incompatible dependencies:

• Verify Python-TensorFlow compatibility: Double-check that your Python version matches the supported range for the TensorFlow CPU version you want to install. For example:
  • TensorFlow 2.15: Python 3.9–3.11
  • TensorFlow 2.16: Python 3.10–3.12
• Install Visual C++ Redistributable: Download and install the latest Microsoft Visual C++ Redistributable for Visual Studio 2019-2022 (x64). This fixes 90% of missing DLL issues for TensorFlow.
• Check for dependency conflicts: Run pip check in your terminal. If any packages are marked as conflicting, uninstall or update them to resolve the issues.

Step 3: Reinstall TensorFlow CPU Properly #

Now let’s do a clean install:

• First, upgrade pip to the latest version:

  python -m pip install --upgrade pip
  

• Install the optimized TensorFlow CPU build (recommended):

  pip install tensorflow-intel
  

  Or install a specific stable version if you need it:

  pip install tensorflow==2.15.0
  

• Verify the installation: Open a Python shell and run:

  import tensorflow as tf
  print(tf.__version__)
  print(tf.config.list_physical_devices('CPU'))
  

  If this runs without errors and shows your CPU device, you’re good to go!

Step 4: Clean Up GPU-Related Environment Variables #

The GPU install likely added CUDA/cuDNN paths to your system environment variables, which can interfere with the CPU version:

• Right-click "This PC" → "Properties" → "Advanced system settings" → "Environment Variables".
• In the "System variables" section:
  • Delete any variables named CUDA_PATH, CUDNN_PATH, or similar.
  • Remove any CUDA/cuDNN-related entries from the Path variable.
• Restart your computer to apply these changes.

If you still run into issues after these steps, try creating a fresh virtual environment for TensorFlow—sometimes old environment conflicts are hard to fully clean up.

内容的提问来源于stack exchange，提问作者netssfy