Ninja error when building PyTorch from sources

#1

Hello there!
I’ve tried to build PyTorch from sources. The following error occurs in the process:

ninja: build stopped: subcommand failed.
Traceback (most recent call last):
  File "setup.py", line 743, in <module>
    build_deps()
  File "setup.py", line 316, in build_deps
    cmake=cmake)
  File "E:\PyTorch\pytorch\tools\build_pytorch_libs.py", line 62, in build_caffe2
    cmake.build(my_env)
  File "E:\PyTorch\pytorch\tools\setup_helpers\cmake.py", line 339, in build
    self.run(build_args, my_env)
  File "E:\PyTorch\pytorch\tools\setup_helpers\cmake.py", line 141, in run
    check_call(command, cwd=self.build_dir, env=env)
  File "C:\Users\Roman\Anaconda3\lib\subprocess.py", line 363, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['cmake', '--build', '.', '--target', 'install', '--config', 'Release', '--', '-j', '4']' returned non-zero exit status 1.

Also i’ve tried this:

git clean -xdf
python setup.py clean
git submodule sync
git submodule deinit -f .
git submodule update --init --recursive

And it didn’t help. An error always occurs in the same place. There are no other errors in the logs.

At the same time a few months ago I’ve successfully assembled PyTorch with the current hardware and software configuration.

What could be the reason?

#2

Hi,

Unfortunately, the true error might be a bit far up in the log. Can you dump all the logs in a text file and share it here?

#3

Do you mean the console output or something else logs?

#4

Console outputs.
Since the build runs in multithreaded way, few things can happen after the first error is detected (we are working on reducing the number of warnings :smiley: )

#5

That’s my outputs: https://gist.github.com/MrLebovsky/6592fe885910f79e5dd3e5e82d3340eb

#6

Thanks !

Could that be an issue with VS v15.9.19 @peterjc123 ?

#7

It was caused by an untested commit to master and it is now reverted.

1 Like
#8

Does it mean that

python setup.py clean
git submodule update --init --recursive

will help?

#9 
python setup.py clean
git pull origin master
git submodule update --init --recursive
#10

Thanks peter!

@MrLebovsky for future reference, note that you can also use the viable/strict branch in the main repo. It contains the latest master for which all the CI passed.

#11

Unfortunately, it does not help. After pull with master I have the same error in different place.
I’ll try to use strict branch.

#12

It helps to resolve my previously problem. But now, after changing the branch to strict , I have the following errors:
https://gist.github.com/MrLebovsky/bd0561c2fda1ef103d3921b809d8e17a

Is it looks like problem with CUDA version?

#13

Here is the status of the CI builds. You may pick one that is all green for Windows.
https://ezyang.github.io/pytorch-ci-hud/build2/pytorch-master

#14

@peterjc123 Building with my computer takes about a day. I’m afraid I didn’t quite understand you and will do something wrong. Could you describe the cause of the problem and its solution in more detail?

As far as I understand, I need to check-out to branch for which all the tests have passed and there will be no errors?
Thanks in advance!

#15

Yeah. I mean that you pick a commit that goes through all the tests and then checkout that commit.
Example:

git clean -xdf
python setup.py clean
git checkout 3c409fc66c76ccbd10bd5d234c6ee8492ae7f653
git submodule sync
git submodule update --init --recursive
#16

@albanD @peterjc123 It does not help :disappointed:
I’ve choouse the commit with all passed tests for windows: 3c409fc66c76ccbd10bd5d234c6ee8492ae7f653
Then I 've launched necessary git commands and the following output’s I have now:

What could be the reason?

#17

Well, I just saw that you are trying to build with CUDA 9.0. I think it is not supported anymore. Only CUDA >= 9.2 builds are well tested now.

#18

And how I can solve it?
My problem is too old hardware with GPU Nvidia GeForce 650M. That’s why I need to build torch from sources.

Can I take some of old commits and try againg or best way is install the CUDA 9.2?

#19

I guess installing CUDA 9.2 is a good choice.