NVIDIA TensorFlow 1.15 Rebuilt

This repository provides a recompiled NVIDIA TensorFlow 1.15.5 pip-package within a Ubuntu 20.04 Docker environment, using CUDA 11.8/12.0/12.1/12.2/12.3, cuDNN 8, and TensorRT 8.

• Without the need to install nvidia-pyindex, this package does not include bundled CUDA runtime installations like nvidia-cuda-runtime-cu12, nvidia-cuda-nvcc-cu12, nvidia-cudnn-cu12, etc. This allows the use of system-level or virtual environment-specific CUDA runtimes, helping to save significant storage space.

• Requires the installation of libnccl2

How to build the pip-wheel:

1. Build the Dockerfile based on the required CUDA version.

2. After starting the built Docker image, the compiled .whl package will be located at /opt/repos/tensorflow/build_items/pip/tensorflow-1.15.5+nv-cp38-cp38-linux_x86_64.whl

   For the compilation process, refer to NVIDIA TensorFlow's Build from Source guide.

---

The compiled pip-wheel package has been placed on the release page.