tensorflow-community-wheels
tensorflow-community-wheels copied to clipboard
yaroslavvb
TensorFlow 2.10.0 Linux GPU (Compute 5.2-8.7), Python 3.8-3.10, CUDA 11.7, cuDNN 8, AVX2, MKL, TensorRT 8
This build solves issues https://github.com/tensorflow/tensorflow/issues/57663 and https://github.com/tensorflow/tensorflow/issues/57671
The wheels are available on the release pages:
- Python 3.8: https://github.com/agkphysics/tensorflow-wheels/releases/tag/tf_2.10.0_gpu_cuda117_cudnn8_avx2_mkl_trt8
- Python 3.9: https://github.com/agkphysics/tensorflow-wheels/releases/tag/tf_2.10.0_gpu_py39_cuda117_cudnn8_avx2_mkl_trt8
- Python 3.10: https://github.com/agkphysics/tensorflow-wheels/releases/tag/tf_2.10.0_gpu_py310_cuda117_cudnn8_avx2_mkl_trt8
@agkphysics This is a welcome contribution, will this be available for Python 3.9 and Python 3.10 (i am interested in the latter) too?
@agkphysics do you have any wheel built with Tensorrt 8.4.3 that align with nividia docker tensorflow 22:08 ?
@agkphysics is it safe (ok) to run this build on an AMD processor? I'm asking because of the MKL part that I suppose is the Intel Math Kernel Library; is this correct?
@fehrin I will try and build with py39 and py310 when I get time.
@dathudeptrai This wheel is built with TensorRT 8.4.3 but I'm not sure how close it is to the nvidia docker container. It says here that tensorflow container version 22.08 has TensorFlow 2.9.1 whereas this wheel is for TensorFlow 2.10.0.
@mariomack I don't see why it won't be safe. However I'm not sure how optimised MKL is for AMD processors (being an Intel product).
@fehrin I've updated the post and added Python 3.9 and Python 3.10 builds.