fo40225/tensorflow-windows-wheel

This repository is a collection of pre-built TensorFlow installation files for Windows. TensorFlow is a machine learning library widely used for building and running AI models. The problem this project solves is that the official TensorFlow releases for Windows are compiled to use AVX instructions, which is a performance feature found in modern CPUs. If your computer has an older or budget CPU that does not support AVX, trying to install the official version will produce a DLL error and TensorFlow simply will not start. The author built alternative versions using SSE2 instead of AVX, which works on a much wider range of hardware. Each release in the table corresponds to a specific combination of TensorFlow version, Python version, whether GPU acceleration is included, and which instruction set the binary targets. GPU-enabled builds also specify the NVIDIA CUDA and cuDNN versions they were compiled against, since those must match what is installed on your machine. To use one of these builds, you find the row in the table that matches your Python version, your CUDA version (if you want GPU support), and your CPU capabilities, then download that file and install it with the pip command using the local filename rather than pulling from the standard package repository. The versions available span from TensorFlow 1.8.0 through 2.9.0, covering Python 2.7 through Python 3.9. The SSE2 variants are specifically the ones to reach for when you see the DLL error on older hardware. The repository does not appear to have received updates beyond version 2.9.0, so it covers the older TensorFlow 1.x and 2.x range rather than the most recent releases.