AttributeError: module 'torchvision.models' has no attribute 'mnasnet'

[anamariaUIC]

🐛 Describe the bug

I installed pytorch via:
conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=9.2 -c pytorch

associated python version is: Python 3.7.4 (default, Aug 13 2019, 20:35:49)
[GCC 7.3.0] :: Anaconda, Inc. on linux

I am running this script: https://github.com/ryujaehun/pytorch-gpu-benchmark/blob/main/benchmark_models.py

and I am getting this Error:
Traceback (most recent call last):
File "python_test_GPU.py", line 22, in
models.mnasnet: models.mnasnet.all[1:],
AttributeError: module 'torchvision.models' has no attribute 'mnasnet'

Versions

Collecting environment information...
PyTorch version: 1.0.1.post2
Is debug build: False
CUDA used to build PyTorch: 9.0.176
ROCM used to build PyTorch: N/A

OS: CentOS Linux 7 (Core) (x86_64)
GCC version: (GCC) 4.8.5 20150623 (Red Hat 4.8.5-28)
Clang version: Could not collect
CMake version: version 2.8.12.2
Libc version: glibc-2.10

Python version: 3.7.4 (default, Aug 13 2019, 20:35:49) [GCC 7.3.0] (64-bit runtime)
Python platform: Linux-3.10.0-862.el7.x86_64-x86_64-with-centos-7.5.1804-Core
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A

Versions of relevant libraries:
[pip3] numpy==1.20.2
[pip3] numpydoc==0.9.1
[pip3] torch==1.0.1.post2
[pip3] torchaudio==0.7.0a0+ac17b64
[pip3] torchvision==0.2.2.post3
[conda] blas 1.0 mkl
[conda] cudatoolkit 10.2.89 hfd86e86_1
[conda] mkl 2019.4 243
[conda] mkl-service 2.3.0 py37he904b0f_0
[conda] mkl_fft 1.0.14 py37ha843d7b_0
[conda] mkl_random 1.1.0 py37hd6b4f25_0
[conda] numpy 1.17.2 py37haad9e8e_0
[conda] numpy-base 1.17.2 py37hde5b4d6_0
[conda] numpydoc 0.9.1 py_0
[conda] pytorch 1.7.0 py3.7_cuda10.2.89_cudnn7.6.5_0 pytorch
[conda] torchaudio 0.7.0 py37 pytorch
[conda] torchvision 0.8.0 py37_cu102 pytorch

[oke-aditya]

I think your pytorch and torchvision versions don't match. Also see #2912

Pytorch = 1.7.0  should have torchvision =  0.8.1 

Edit:
Also you have conflicting conda and pip installation.
The best suggested way is to install using only conda as per the getting started guide

[oke-aditya]

You should try

conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=9.2 -c pytorch

[anamariaUIC]

@oke-aditya Thanks how do I uninstall the previous version that I installed? Or I don't need to do anything just run that command that you provided?

[oke-aditya]

You can either create a fresh environment using conda and install again. (easy way)

Or uninstall packages using pip and conda then install

[anamariaUIC]

@oke-aditya can you please send me command how do I uninstall it using pip in order to remove all files?
Do I only do:
pip uninstall torch

Please send me all commands for the complete uninstallation

[oke-aditya]

I would suggest to remove all the packages.

pip uninstall numpy numpydoc torch torchaudio torchvision

(Press y to say yes and remove stuff)

Then cleanup the conda environment. (Coz your device is CUDA 9.2 and you have CUDA 10. 2 installed so pytorch won't run)

This will activate the base environment

conda activate base

Beware!! The below command will completely remove the environment.

conda env remove -n YOUR_PREVIOUS_ENV_NAME

Now you can create a fresh environment

conda create --name NEW_ENVIRONMENT_NAME

conda activate NEW_ENVIRONMENT_NAME

And proceed with installation as mentioned on the site.

Hope I helped you. :) It's bit hard to communicate installation issues :(

[anamariaUIC]

@oke-aditya Thank you so much for your help. I tried installing with the command you gave me:
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=9.2 -c pytorch

but when I run this

I get this:
Traceback (most recent call last):
File "python_test_GPU.py", line 22, in
models.mnasnet: models.mnasnet.all[1:],
AttributeError: module 'torchvision.models' has no attribute 'mnasnet'
end

vi system_info.txt
benchmark start : 2021/12/11 14:34:46
system_configs

uname_result(system='Linux', node='gpu-2-0.saber', release='3.10.0-862.el7.x86_64', version='#1 SMP Fri Apr 20 16:44:24 UTC 2018', machine='x86_64', processor='x86_64')
scpufreq(current=2100.0, min=0.0, max=0.0)
cpu_count: 16
memory_available: 90282246144
gpu_configs

Number of GPUs on current device : 1
CUDA Version : 9.0.176
Cudnn Version : 7402
Device Name : Tesla V100-PCIE-16GB
benchmark end : 2021/12/11 14:58:58

Can you please advice how to install pytorch so that the above scrip can run, this module part:
MODEL_LIST = {
models.mnasnet: models.mnasnet.all[1:],
models.resnet: models.resnet.all[1:],
models.densenet: models.densenet.all[1:],
models.squeezenet: models.squeezenet.all[1:],
models.vgg: models.vgg.all[1:],
models.mobilenet: models.mobilenet.mv2_all[1:],
models.mobilenet: models.mobilenet.mv3_all[1:],
models.shufflenetv2: models.shufflenetv2.all[1:],
}

For this version of pytorch installation these modules don't work:
models.mnasnet
models.mobilenet
models.mobilenet

What is the version and how to install Pytorch so that it does work with my:
Cuda 9.2
Drivers: NVIDIA-SMI 396.44, Driver Version: 396.44

Please advise,
Thanks!

[pmeier]

@anamariaUIC It should be one of

• mnasnet0_5
• mnasnet0_75
• mnasnet1_0
• mnasnet1_3

See this list for all available classification models.

[anamariaUIC]

@pmeier @oke-aditya thank you for getting back to me. @pmeier How should I use the information from your comment to either install or run pytorch, given I want to run this script

[NicolasHug]

@anamariaUIC please follow the latest pytorch version following the official instructions https://pytorch.org/get-started/locally/

FYI this works fine (from the script you linked) with the latest version:

MODEL_LIST = {
    models.mnasnet: models.mnasnet.__all__[1:],
    models.resnet: models.resnet.__all__[1:],
    models.densenet: models.densenet.__all__[1:],
    models.squeezenet: models.squeezenet.__all__[1:],
    models.vgg: models.vgg.__all__[1:],
    models.mobilenet: models.mobilenet.mv2_all[1:],
    models.mobilenet: models.mobilenet.mv3_all[1:],
    models.shufflenetv2: models.shufflenetv2.__all__[1:],
}

Since this is a usage question more than a bug or an issue related to torchvision, I'll close the issue.

[pmeier]

That is a question you should ask the author of the script. You can import the MNASNet models like this:

from torchvision import models

models.mnasnet0_5(...)

[anamariaUIC]

@NicolasHug does the latest version of pytorch works with CUDA 9.2? Because if it doesn't I can't use it. I am doing this on a cluster where drivers are version 396 (please see details from my previous comments) and only Cuda 9.2 can be used...

[NicolasHug]

@anamariaUIC please see https://pytorch.org/get-started/previous-versions/ for cuda / pytroch compatibility and our readme for torch / torchvision compatibility.

There's a chance that not all models will be available in torchvision if you need to rely on such an old CUDA version, unfortunately

[pmeier]

does the latest version of pytorch works with CUDA 9.2?

Nope. The last release that supports CUDA 9.2 is torch==1.7.1 / torchvision==0.8.2.

[anamariaUIC]

ok, So I did install the latest version of pytorch that works with CUDA 9.2 via:

conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=9.2 -c pytorch

And for example models.mnasnet doesn't work with it as invoked in this script:
models.mnasnet: models.mnasnet.all[1:]

Is this only the issue of invoking it instead with:
models.mnasnet: models.mnasnet0_5.all[1:],

or mnasnet: models doesn't work at all with pytorch==1.7.1?