Error installing current nightly (dev20210122)

1

Hello Forum!

Conda install on linux of current nightly (1/22/21) is failing. Is this just me,
or is there a systematic issue? (I was just able to install the corresponding
stable, 1.7.1.)

Here is my install command:

conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch-nightly

And here is the error message:

Downloading and Extracting Packages
pytorch-1.8.0.dev202 | 873.5 MB  |                                         |   0%

InvalidArchiveError('Error with archive /home/user/miniconda3/pkgs/pytorch-1.8.0.dev20210122-py3.8_cuda11.0.221_cudnn8.0.5_0.tar.bz2.  You probably need to delete and re-download or re-create this file.  Message from libarchive was:\n\nSeek failed')

I did try deleting the specified file, and reinstalling, but got the same error.

Thanks.

K. Frank

2

Hi KFrank,

thanks for reporting this issue. I seem to run into the same error and I guess the download is failing somehow.

Downloading and Extracting Packages
lz4-c-1.9.3          | 186 KB    | ######################################################################################################################################################################## | 100%
pytorch-1.8.0.dev202 | 873.5 MB  |                                                                                                                                                                          |   0%
torchaudio-0.8.0.dev | 14.2 MB   | ######################################################################################################################################################################## | 100%
torchvision-0.9.0.de | 26.2 MB   | ######################################################################################################################################################################## | 100%

InvalidArchiveError('Error with archive .../pytorch-1.8.0.dev20210122-py3.8_cuda11.0.221_cudnn8.0.5_0.tar.bz2.  You probably need to delete and re-download or re-create this file.  Message from libarchive was:\n\nSeek failed')

The file itself is available here and wget yields a slow download speed of ~500kB/s.
I’m unsure, if conda uses some timeouts or what the reason is.

@seemethere do you have an idea and can reproduce it?

PS: I’m using a new conda env with Python=3.8.

UPDATE:
downloading the file manually and installing it via conda install ./pytorch-1.8..... also fails with the same error, so it doesn’t seem to be related to a timeout.
Unpacking the .tar.bz also works fine without an issue using tar xvf.

UPDATE2:
It seems to be a known issue in conda-package-handling with a potential PR.

3

Hi @ptrblck!

Thanks for checking on this. This is not urgent for me – I have a modestly
older nightly build installed (also 1.8), if I want to play around.

I just like checking up on the nightly build from time to time. You know …
Just to make sure that the devs aren’t getting up to any shenanigans!

Best.

K. Frank

1 Like
4

Hahaha, thanks for keeping an eye on this! :wink:

We would have to take another look into this issue, as the nightlies are especially useful to test new features or debug new issues.

5

Same here I guess, but it only shows no progress in download?

image
image1919Ă—243 26.7 KB

6

Hi @ptrblck!

Just FYI: I tried installing the nightly again (on 2/2/21). It still fails, but with a
different kind of error, maybe about my python version:

(goofy) user@server:~$ conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch-nightly
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: -
Found conflicts! Looking for incompatible packages.                             failed

UnsatisfiableError: The following specifications were found
to be incompatible with the existing python installation in your environment:

Specifications:

  - olefile -> python[version='>=2.7,<2.8.0a0|>=3.5,<3.6.0a0|>=3.6,<3.7.0a0|>=3.7,<3.8.0a0']
  - torchvision -> python[version='>=3.9,<3.10.0a0']
  - typing_extensions -> python[version='>=3.9,<3.10.0a0']
  - wheel -> python[version='>=3.9,<3.10.0a0']

Your python: python=3.8

(I could post the entire list of conflicts, if that would be helpful. But it’s long.)

Has the nightly now changed to requiring python 3.9? The install instructions
on the main pytorch.org web page seem to imply that 3.9 is not yet required:

NOTE: Python 3.9 users will need to add '-c=conda-forge' for installation
conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch-nightly

Again, this is not important to me – I’m just providing feedback.

Best.

K. Frank

1 Like
7

I also encountered the same conflicts errors. there is a solution for nightly install?

Thanks.

8

Hi Natan!

I do not understand how conda works, so I don’t know what I’m doing.

However …

I’ve tried from time to time to use conda to install the nightly build (on
linux) and kept getting the “Solving environment: failed with initial
frozen solve.” set of errors.

So I tried creating a new conda environment, and got the same error.
Then I tried creating a new conda environment and cleaning it before
trying the install (“conda clean -a”).

Then I ran the install:

conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch-nightly

This time the install succeeded:

>>> import torch
>>> torch.__version__
'1.9.0.dev20210213'

Why I would need to “clean” a newly-created environment to get the
install to work, I have no idea. But that’s what finally worked for me.

Best.

K. Frank

9

I am still having this issue. The download stalls during conda install.

.

10

For 3090 the current most up to date for 1.9 is install it with

pip install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu111/torch_nightly.html

Hope this is helpful.