U20.4 升级 pytorch 1.11

一 系统环境

1，系统已安装 pytorch 1.7 , 1.8

a)

pip install torch==1.8.2+cu111 torchvision==0.9.2+cu111 torchaudio===0.8.2 -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html

或者
https://download.pytorch.org/whl/torch_stable.html
cu111/torch-1.8.1%2Bcu111-cp38-cp38-linux_x86_64.whl
cu111/torchvision-0.9.1%2Bcu111-cp38-cp38-linux_x86_64.whl

$nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Mon_Nov_30_19:08:53_PST_2020
Cuda compilation tools, release 11.2, V11.2.67
Build cuda_11.2.r11.2/compiler.29373293_0

b) 因为时间久远忘记是安装的哪一个了，可以自己去NVIDIA

cudnn-11.3-linux-x64-v8.2.1.32.tgz 或 NVIDIA-Linux-x86_64-460.91.03.run

cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
#define CUDNN_MAJOR 8
#define CUDNN_MINOR 2
#define CUDNN_PATCHLEVEL 1
--
#define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

#endif /* CUDNN_VERSION_H */

2 升级 

2.1 升级nvidia-driver

$nvidia-smi
Tue May 17 10:22:01 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.60.02    Driver Version: 510.60.02    CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0 Off |                  N/A |
| 40%   68C    P2   200W / 280W |   7891MiB /  8192MiB |     97%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

$ubuntu-drivers

== /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0 ==
modalias : pci:v000010DEd00002488sv00001462sd00003901bc03sc00i00
vendor   : NVIDIA Corporation
driver   : nvidia-driver-470 - distro non-free recommended
driver   : nvidia-driver-510-server - distro non-free
driver   : nvidia-driver-510 - distro non-free
driver   : nvidia-driver-470-server - distro non-free
driver   : xserver-xorg-video-nouveau - distro free builtin

然后，可以直接apt install nvidia-driver-510 , 或者

在 软件和更新 中选择 附加驱动，然后选择推荐的驱动，点击 应用更改

2.2 创建torch 1.11环境

$conda create -n aigret python=3.10 ipykernel psutil jupyter jupyterlab nodejs numpy matplotlib

环境名称 aigret， python版本3.10 ,后面是一些会用的的软件，直接一起安装了。

$conda activate aigret
$conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    blas-1.0                   |              mkl           6 KB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
    cudatoolkit-11.3.1         |       h2bc3f7f_2       549.3 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    ffmpeg-4.3                 |       hf484d3e_0         9.9 MB  pytorch
    gmp-6.2.1                  |       h58526e2_0         806 KB  conda-forge
    gnutls-3.6.13              |       h85f3911_1         2.0 MB  conda-forge
    intel-openmp-2022.0.1      |    h06a4308_3633         4.2 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    lame-3.100                 |    h7f98852_1001         496 KB  conda-forge
    libblas-3.9.0              |   14_linux64_mkl          13 KB  conda-forge
    libcblas-3.9.0             |   14_linux64_mkl          12 KB  conda-forge
    liblapack-3.9.0            |   14_linux64_mkl          12 KB  conda-forge
    mkl-2022.0.1               |     h06a4308_117       127.7 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    nettle-3.6                 |       he412f7d_0         6.5 MB  conda-forge
    openh264-2.1.1             |       h780b84a_0         1.5 MB  conda-forge
    pytorch-1.11.0             |py3.10_cuda11.3_cudnn8.2.0_0        1.02 GB  pytorch
    pytorch-mutex-1.0          |             cuda           3 KB  pytorch
    torchaudio-0.11.0          |      py310_cu113         5.3 MB  pytorch
    torchvision-0.12.0         |      py310_cu113        27.5 MB  pytorch
    typing_extensions-4.2.0    |     pyha770c72_1          27 KB  conda-forge
    ------------------------------------------------------------
                                           Total:        1.74 GB

然后在jupyterlab上选择环境并验证可用性。有意思的是这里不是显示1.11.0+cu113，而是如下：

import torch
print(torch.__version__)
print(torch.cuda.is_available())

====
1.11.0
True
====
1.8.2+cu111
True

 代码没有任何变动，可以正常运行。后续研究一下有什么新功能，可以提供速度。