Hey there, let's break down why this error is happening and walk through fixes tailored to your setup!

Why the Error Occurs #

The core issue here is version compatibility: PyTorch 1.4.0 launched before Python 3.8 was released, so there are no prebuilt installation packages (wheels) for Python 3.8. That's exactly why pip can't find a matching version—those packages simply don't exist for your Python setup.

As for whether others can use that command: Yes, folks running Python 3.6 or 3.7 (the officially supported versions for PyTorch 1.4.0) would execute that pip command without issues. Your pip itself is working fine (since you can install other modules), this is just a mismatch between the PyTorch version you're targeting and your Python version.

Fixes to Try #

Here are practical solutions for your Windows 10, CUDA 10.1 environment:

• Option 1: Downgrade Python to 3.7
  If you're set on using PyTorch 1.4.0 specifically, downgrade your Python to 3.7 (a version fully supported by PyTorch 1.4.0). I recommend using a virtual environment (like venv or conda) to do this—you'll avoid disrupting your existing Python 3.8 setup.

• Option 2: Upgrade PyTorch to a Python 3.8-compatible version
  If you want to keep Python 3.8, switch to a newer PyTorch release that adds support for it. For CUDA 10.1, versions 1.5.0 and above work with Python 3.8. A stable, reliable choice is:

  pip install torch==1.7.1 torchvision==0.8.2
  

  This version pair is fully compatible with your Python 3.8 and CUDA 10.1 setup.

• Verify your CUDA configuration
  Just to be safe, confirm your CUDA 10.1 is properly installed and detected. Run this command in your command prompt:

  nvcc --version
  

  It should output CUDA 10.1 version details—if not, you might need to adjust your system's PATH variables to include the CUDA binaries.

内容的提问来源于stack exchange，提问作者Allie