Pytorch 0.3 version is released


(Mu Wu9527) #1

Pytorch 0.3 version is released. What new features is added in this version?


#2

Probably, this is what you want.


(Royi) #3

I hope Anaconda will update its package fast.
Speed gains are always welcome :-).


#4

You don’t need to install from the anaconda package. The following will work (if you’re running cuda8.0 or no cuda):
conda install pytorch torchvision -c pytorch

See http://pytorch.org/ for more details


(Xiaotao Gu) #5

Hi @richard! I just tried to install version 0.3.0 from Conda following the above instruction, but it turned out that the installed version is 0.1.12. For detailed information, I am running this on Linux with up-to-date Conda. Any idea about this issue? Thanks!


(Simon Wang) #6

I just tried the command it finds 0.3 for me on Ubuntu, CentOS and OSX. What distro are you on?


(jdhao) #7

Same for CentOS. We can use conda install pytorch torchvision -c pytorch to update torch and torchvision to the latest version (0.3.0 and 0.2.0 respectively).


(Xiaotao Gu) #8

Surprisingly it works now :sunny: Thanks guys!


(Royi) #9

I prefer using Anaconda Navigator.
It seems that now they updated the official package of PyTorch in their official channel.
So it should be available on Anacaonda Navigator.

Thank You.


#10

hi, when I use “conda install pytorch torchvision -c pytorch” command and I get this

sunhanbo@dg33:~/MOT/LMP$ conda install pytorch torchvision -c pytorch
Fetching package metadata …
Solving package specifications: .

Package plan for installation in environment /home/sunhanbo/anaconda3:

The following NEW packages will be INSTALLED:

pytorch:     0.1.12-py36cuda8.0cudnn6.0_1 https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
torchvision: 0.1.8-py36_0                 https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free

Proceed ([y]/n)? n

what can I do to get pytorch 0.3.0


#11

Try to reinstall/update anaconda with the latest version. It helped some people with your problem.


(Simon Wang) #12

It looks like you are behind a proxy? The conda installation url is hosted by tsinghua.


(jdhao) #13

I upgraded to the latest torch and torchvision using conda install pytorch torchvision -c pytorch as suggest by @richard. I have notice that originally we install these 2 packages from the soumith channel. Now I try to install graphviz python package, condat tries to downgrade torch and torchvision. The complete message is shown below,

[~]$ conda install graphviz
Fetching package metadata …
Solving package specifications: .

Package plan for installation in environment /home/jdhao/util/anaconda3:

The following NEW packages will be INSTALLED:

graphviz: 2.38.0-5
libtool: 2.4.6-h544aabb_3
pango: 1.40.3-1

The following packages will be UPDATED:

cairo: 1.14.6-4 conda-forge --> 1.14.8-0
qt: 5.6.2-3 conda-forge --> 5.6.2-5

The following packages will be SUPERSEDED by a higher-priority channel:

fontconfig: 2.12.1-4 conda-forge --> 2.12.1-3
freetype: 2.7-1 conda-forge --> 2.5.5-2
glib: 2.51.4-0 conda-forge --> 2.50.2-1
gstreamer: 1.8.0-2 conda-forge --> 1.8.0-0
harfbuzz: 1.4.3-0 conda-forge --> 0.9.39-2
matplotlib: 2.1.0-py36_0 conda-forge --> 2.0.2-np113py36_0
pillow: 4.3.0-py36_0 conda-forge --> 4.2.1-py36_0
pytorch: 0.3.0-py36_cuda8.0.61_cudnn7.0.3h37a80b5_4 pytorch --> 0.2.0-py36h53baedd_4cu80 soumith [cuda80]
torchvision: 0.2.0-py36h17b6947_1 pytorch --> 0.1.9-py36h7584368_1 soumith

The following packages will be DOWNGRADED:

icu: 58.2-h211956c_0 --> 54.1-0

Proceed ([y]/n)?

How to resolve this? Just remove the soumith channel?


Updating PyTorch
(jdhao) #14

I remove the soumith channel using conda config --remove channels 'soumith' and then installed graphviz. It seems there is no error.


(Jorge Balazs) #15

Upgrading anaconda did it for me. With no env active, I ran:

conda upgrade conda
source activate <env_name>
conda install pytorch torchvision -c pytorch

And it listed the right versions.


(Adarsh Kumar) #16

Instead of upgrading conda, I tried pip and it worked for me.

Command : pip install http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp35-cp35m-linux_x86_64.whl


#17

You can add other proxy servers.

conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/free/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/main/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/conda-forge/
conda install pytorch torchvision -c pytorch

It works for me.