求适配CUDA 9.1的TensorFlow预编译GPU版本及编译问题求助

Hey there, I totally get how frustrating this is—compiling TensorFlow from scratch is a total slog, especially when you're new to the ecosystem and just want to get up and running with your MX150. Let’s walk through your options to get a working pre-built version:

• First, lock in compatible TensorFlow versions: TensorFlow 1.9 through 1.12 officially support CUDA 9.1. These are the versions you’ll want to target, as they have pre-built GPU packages tailored for this CUDA release.
• Grab official pre-built .whl packages: Head to the TensorFlow GitHub releases page (no external links needed—just search for "TensorFlow releases" and filter for versions 1.9–1.12). Look for files named like tensorflow_gpu-1.12.0-cp37-cp37m-win_amd64.whl (adjust the Python version and OS to match yours). Once downloaded, install it with:

  pip install path/to/your/tensorflow_gpu_package.whl
  

• Community-shared builds if official ones don’t fit: If you need a specific Python/OS combination not covered by official builds, check community platforms where developers share their own compiled TensorFlow packages. Stick to trusted sources to avoid issues, and double-check that the build explicitly lists CUDA 9.1 compatibility.
• Docker as a hassle-free alternative: If downloading a whl feels tricky, use a pre-built Docker image that bundles TensorFlow and CUDA 9.1. Look for tags like tensorflow/tensorflow:1.11.0-gpu—these images are pre-configured, so you can start using TensorFlow without any setup beyond running the container.

Once you install the right package, you can verify it’s using your GPU by running this quick check in Python:

import tensorflow as tf
print(tf.test.is_gpu_available())

It should return True if everything’s set up correctly.

Good luck—you’ll be training models on your MX150 in no time!

内容的提问来源于stack exchange，提问作者Pierre Carceller