Vision Transformer
Tensorflow-GPU版本与CUDA版本对应关系
tensorflow版本 | Python版本 | cuDnn | CUDA |
---|
tensorflow_gpu-2.6.0 | 3.6-3.9 | 8.1 | 11.2 | tensorflow_gpu-2.5.0 | 3.6-3.9 | 8.1 | 11.2 | tensorflow_gpu-2.4.0 | 3.6-3.8 | 8.0 | 11.0 | tensorflow_gpu-2.1.0~2.3.0 | 3.5-3.8 | 7.6 | 10.1 | tensorflow_gpu-2.0.0 | 3.5-3.7 | 7.4 | 10 | tensorflow_gpu-1.13.0~1.15.0 | 3.5-3.7 | 7.4 | 10 | tensorflow_gpu-1.5.0~1.11.0 | 3.5-3.6 | 7 | 9 |
Vision Transformer使用到了MultiHeadAttention,对应tensorflow-gpu最低版本为2.4.0
安装tensorflow-gpu版本
conda install cudatoolkit=11.0
conda install cudnn==8.0.5.39 -c conda-forge
pip install tensorflow==2.4.0
指定pip源
pip install xxx -i https://mirrors.aliyun.com/pypi/simple/
pip install xxx -i https://mirrors.tencent.com/pypi/simple/
pip install xxx -i https://pypi.douban.com/simple/
pip install xxx -i https://pypi.tuna.tsinghua.edu.cn/simple/
Vision Transformer tensorflow实现
|