安装cdqa时遇torch==1.2.0版本匹配错误，求解决方法

Yep, I've dealt with this exact headache before when setting up cdqa! The core issue is that cdqa has a hard dependency on an older version of PyTorch (1.2.0), which isn't available via standard PyPI for newer Python versions or some system architectures. Here's how to fix it:

1. Verify Your Python Version First #

Torch 1.2.0 only supports Python 3.6 or 3.7 — it won't work with Python 3.8 or later. Run this command to check your current version:

python --version

If you're on a newer Python version, create a virtual environment with Python 3.7 to avoid breaking other projects. For example, using conda:

conda create -n cdqa_env python=3.7
conda activate cdqa_env

Or with venv if you prefer:

python3.7 -m venv cdqa_env
source cdqa_env/bin/activate  # On macOS/Linux
cdqa_env\Scripts\activate     # On Windows

2. Install torch==1.2.0 Directly from PyTorch's Repository #

Even with the right Python version, standard PyPI might not have the wheel for your system. Install torch 1.2.0 using this command (it pulls from PyTorch's official wheel store):

pip install torch==1.2.0 -f https://download.pytorch.org/whl/torch_stable.html

Once torch is installed, you can then safely run:

pip install cdqa

3. Fallback: Install cdqa Without Strict Dependencies (Risky but Functional) #

If you can't use Python 3.7 (e.g., your system doesn't support it), you can install cdqa without enforcing its dependency constraints, then install a newer compatible torch version. Note: This might cause some minor compatibility issues, but it works for most core features.

First, install cdqa without dependencies:

pip install cdqa --no-deps

Then install the required dependencies manually (use a recent torch version that works with your Python):

pip install torch transformers==2.8.0 pandas numpy scikit-learn

That should get you up and running with cdqa without the version mismatch error!

内容的提问来源于stack exchange，提问作者roney