使用PIP安装TensorFlow 1.15.0 GPU版本失败的技术求助

Hey there, let's work through this TensorFlow 1.15.0 GPU installation issue you're hitting. The error you're seeing—where pip can't find the 1.15.0 version but works fine with newer TensorFlow releases—almost always boils down to compatibility mismatches or configuration oversights. Here's how to fix it step by step:

1. Verify Your Python Version Compatibility #

TensorFlow 1.15.0 only supports Python 2.7, 3.5, 3.6, or 3.7. If you're running Python 3.8 or newer, pip won't be able to locate a matching distribution for TensorFlow 1.x.

• Check your current Python version with:

  python --version  # or python3 --version
  

• If your version is too new, install a compatible Python release (e.g., Python 3.7). You can use tools like pyenv for easy version management, or download the installer directly from the official Python website.

2. Downgrade pip to a Compatible Version #

Newer pip versions (21.x and above) have dropped support for some older package indexes used by TensorFlow 1.x. Downgrading to a slightly older pip version often resolves this "no matching distribution" error.

Run this command to downgrade pip:

pip install pip==20.3.4

3. Install the Correct GPU-Specific Package #

Unlike TensorFlow 2.x (which combines CPU and GPU support in one package), TensorFlow 1.x separates GPU support into a distinct package. You need to install tensorflow-gpu==1.15.0 instead of the generic tensorflow==1.15.0.

Use this command (add a PyPI mirror if needed):

pip install tensorflow-gpu==1.15.0

4. Ensure CUDA and cuDNN Version Match #

TensorFlow 1.15.0 requires specific versions of NVIDIA's CUDA and cuDNN libraries to work properly:

• CUDA 10.0 (not 10.1, 11.x, or newer)

• cuDNN 7.6.5 (compatible with CUDA 10.0)

• Download these versions from NVIDIA's official site

• Configure environment variables to point to the CUDA installation:

  • For Windows: Add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin and C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\libnvvp to your PATH.
  • For Linux/macOS: Add these lines to your shell config (e.g., ~/.bashrc or ~/.zshrc):

    export CUDA_PATH=/usr/local/cuda-10.0
    export PATH=$CUDA_PATH/bin:$PATH
    export LD_LIBRARY_PATH=$CUDA_PATH/lib64:$LD_LIBRARY_PATH
    

5. Use a Domestic PyPI Mirror (If Applicable) #

If you're located in a region with limited access to the official PyPI servers, using a domestic mirror can help resolve package discovery issues. For example, use the Tsinghua University mirror:

pip install tensorflow-gpu==1.15.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

内容的提问来源于stack exchange，提问作者Deepak Raj