pip install error #1
Comments
|
Please give the full output of error message, and make sure you are using |
|
does it work when compiling from source ? if so, can you try |
|
this is my full error message when try
Collecting spatial-correlation-sampler
and this is the message that i tried
running install thanks. |
|
yes i'm using torch version 0.4.1 |
|
what is your cuda version ? and your gcc version ? the error seems to be in nvcc. |
|
i installed cuda version 9.0.176 and gcc version 6.4.0 |
|
can you try with python3 ? Just to know if nvcc is the problem or my code (If so, I'll try to make my code more compatible with python2) |
|
ok i’ll try with python3 |
|
tried to install both ways, 1) pip 2) git clone and setup py. Seems like users over at the Nvidia page get similar error messages. |
|
Hello, same as above, can you state your python version, pytorch version, cuda version and GCC version ? |
|
3.6, 4.1, V8.0.61, 4.8.4 |
|
Unfortunately, I don't know what is wrong with your setup. Best usecase would be to test it in a docker with your your setup, but I don't have much time for now :( Best advice (although admittedly not sufficient is to try with more up to date gcc and CUDA) would be use more up to date cuda and gcc version Also, make sure your pytorch is compiled with the same CUDA version as the one installed in your computer. |
|
@ClementPinard What is the cuda, ubuntu, and gcc version on which the test passes? |
|
Tested with Ubuntu 16.04, cuda 9.2 and GCC 5. Do you experience similar problems for your setup? |
|
I tried to set up with Ubuntu 18.04, cuda 9.0 and gcc 6. The problem is nvcc 9.0 is not compatible with gcc>5. And if I compile your package with gcc5 using, Bottomline: Switch to cuda 9.2 which supports gcc >=5. Also, cuda 10.0 has been released, so there are going to be more problems. |
|
I am not very familiar with gcc4/5 differences, but wouldn't the problem be also present in the custom CPP extension official pytorch tutorial? Did this repo work for you ? |
|
I recall being able to use it with 9.1 but didn't test 9.0 though |
|
I din't try the https://github.com/pytorch/extension-cpp on cuda 9.0., and will avoid going back to it. Later, pytorch's ATen calls include headers from |
|
May be, we should add it to readme. The following works/ doesnt work with pytorch 0.4: Works: If you have both |
|
I updated the README, thanks for your work on compatibility for this module ! Hopefully, pytorch CUDAExtension module will be able to deal properly with such CUDA/gcc incompatibilities in the future. |
|
Hey, I had to build this module on a system with CUDA 9.0 and gcc-7 by default
|
|
Yes, this sounds like reasonable fix. Although, I don't have sudo access on my gpu machines. :D |
|
I use the following scripts to switch to another gcc version and install the correlation lib successfully on CUDA 9.2 (original gcc-v8 not work) |
|
@charliememory can you please tell me how did you get gcc-5.4.0 on your machine? I have gcc-4.8.5. and I don't have sudo access so I think I won't be able to install it, although your method of switching to another version seems like the only choice I have. |
|
@moha23 You can install it without sudo. While you need to make and compile from the gcc source code. |
|
Maybe you can install it from anaconda ? |
|
@ClementPinard I installed gcc7.3 using Conda (helpful StackOverflow Post) and still getting error when trying to pip install. Python 3.7.7, pytorch 1.5.0, cuda10.0(server has both cuda9 and cuda10) and GCC 7.3 Could it possibly be because of Conda installed GCC? They have different names but according to https://docs.conda.io/projects/conda-build/en/latest/resources/compiler-tools.html environment variables are set to point to the correct compiler. EDIT: I just checked pytorch has the compiler names hardcoded, and this function checks for it. Probably this is what is causing the error. Error: ERROR: Command errored out with exit status 1:
|
|
Can you put the log when you install it directly from source instead ? Other than that, pytorch gets the compiler that the environment variable
|
|
Installing directly from source gives: which: no hipcc in (/home/s18002/anaconda3/envs/try1/bin:/home/s18002/anaconda3/condabin:.local/bin:/opt/soft/share/cudann-10.0/bin:/opt/soft/share/cuda-10.0/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/home/s18002/.local/bin:/home/s18002/bin) |
This comment has been minimized.
This comment has been minimized.
|
I just built gcc7.3.0 and it got installed all good. |
|
I built gcc7.3.0 ,but it didnt work ,too. which command do you input after built gcc7.3.0? |
|
@charliememory should i input your scripts in ~/.bashrc and source ~/.bashrc? |
|
@moha23 hi! I have a same problem, I installed gcc7.3.0 using Conda. |
Hi, I don't really remember, but as per my comments I probably built gcc from source and not conda. Sorry can't be of much help. |
|
okay，I try to find other solutions，thank you！
…------------------ 原始邮件 ------------------
发件人: "ClementPinard/Pytorch-Correlation-extension" ***@***.***>;
发送时间: 2021年6月9日(星期三) 晚上8:42
***@***.***>;
***@***.******@***.***>;
主题: Re: [ClementPinard/Pytorch-Correlation-extension] pip install error (#1)
@moha23 hi! I have a same problem, I installed gcc7.3.0 using Conda.
pytorch 1.0 cuda 10.0 gcc7.3.0
Hi, I don't really remember, but as per my comments I probably built gcc from source and not conda. Sorry can't be of much help.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or unsubscribe.
|
hi i failed to install
i got this error message
Command "/usr/bin/python -u -c "import setuptools, tokenize;file='/tmp/pip-build-QlQ2v1/spatial-correlation-sampler/setup.py';f=getattr(tokenize, 'open', open)(file);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, file, 'exec'))" install --record /tmp/pip-1qIZX1-record/install-record.txt --single-version-externally-managed --compile" failed with error code 1 in /tmp/pip-build-QlQ2v1/spatial-correlation-sampler/
i don't even know what this mean please help me
The text was updated successfully, but these errors were encountered: