error in one time setup

[TimSousa1]

when I run: pip install --global-option="--no-networks" git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch i get

    ERROR: Command errored out with exit status 1:
     command: 'C:\Users\nunob\anaconda3\envs\dmodel\python.exe' -c 'import io, os, sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\\Users\\nunob\\AppData\\Local\\Temp\\pip-req-build-ajsps1vv\\bindings/torch\\setup.py'"'"'; __file__='"'"'C:\\Users\\nunob\\AppData\\Local\\Temp\\pip-req-build-ajsps1vv\\bindings/torch\\setup.py'"'"';f = getattr(tokenize, '"'"'open'"'"', open)(__file__) if os.path.exists(__file__) else io.StringIO('"'"'from setuptools import setup; setup()'"'"');code = f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' egg_info --egg-base 'C:\Users\nunob\AppData\Local\Temp\pip-pip-egg-info-gm8vy1dn'
         cwd: C:\Users\nunob\AppData\Local\Temp\pip-req-build-ajsps1vv\bindings/torch
    Complete output (7 lines):
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "C:\Users\nunob\AppData\Local\Temp\pip-req-build-ajsps1vv\bindings/torch\setup.py", line 3, in <module>
        import torch
      File "C:\Users\nunob\anaconda3\envs\dmodel\lib\site-packages\torch\__init__.py", line 126, in <module>
        raise err
    OSError: [WinError 182] The operating system cannot run %1. Error loading "C:\Users\nunob\anaconda3\envs\dmodel\lib\site-packages\torch\lib\shm.dll" or one of its dependencies.

[jmunkberg]

This seems to be a problem in the tiny-cuda-nn setup. I would suggest reporting that issue there https://github.com/NVlabs/tiny-cuda-nn, or testing their bindings in isolation to track down the issue: https://github.com/NVlabs/tiny-cuda-nn#pytorch-extension

[MR-JR-01]

I got the one time setup to work on the thousandth try (because im not a programmer/scientist, just very stubborn) - so I hope this will help some people - heres a sketchfab model of bob so you can see I actually managed to get this working

I found that the dmodel environment wasnt working, but found it works in the torch environment we're going to create here.

Before we begin

your GPU must be NVIDIA CUDA enabled - check yours here https://developer.nvidia.com/cuda-gpus - keep the page open as we'll need this later If you do not know what GPU you have go to Start>Device Manager and look under 'display adapters' then compare this to the nvidia cuda gpus page.

Important!

ensure you have set this environment variable up:

in the System variables click 'new' and in variable name enter TCNN_CUDA_ARCHITECTURES in variable value enter your GPU compute value - this can be found on NVIDIA https://developer.nvidia.com/cuda-gpus e.g. mine has a Quadro RTX 4000 which has a compute capability of 7.5 - hint - remove the full stop - so 7.5 becomes 75, enter the latter: 75 as the value.

Lets go:

create the conda environment

conda create -n torch python=3.6

activate the created environment

conda activate torch

install numpy (if this returns 'requirement already satisfied' then move on to next step)

pip install numpy

install pytorch (I recommend the below, not tested with any other versions)

conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge

Test GPU install

python -c 'import torch; print(torch.rand(2,3).cuda())'

this should return something like:

tensor([[0.5690, 0.3455, 0.4558], [0.7590, 0.6625, 0.0733]], device='cuda:0')

thanks (https://github.com/williamFalcon/pytorch-gpu-install)

double check GPU install - in conda type:

python

(this will enable the python interpreter - you can see this is active as the line now starts with >>> then copy the bullet pointed text only

• import torch

• torch.cuda.is_available()

this should then return True or False - it should return True following this method so far

• torch.cuda.device_count()

this should return the number of devices, in my case: 1

• torch.cuda.current_device()

0

• torch.cuda.device(0)

<torch.cuda.device at 0x7efce0b03be0>

• torch.cuda.get_device_name(0) 'Quadro RTX 4000'

exit python interpreter

• quit()

then install following from nvdiffrec instructions

pip install ninja imageio PyOpenGL glfw xatlas gdown

pip install git+https://github.com/NVlabs/nvdiffrast/

now do this (this is where most of us have been having problems with bindings)

pip install --global-option="--no-networks" git+https://github.com/NVlabs/tiny-cuda-nn#subdirectory=bindings/torch

then

imageio_download_bin freeimage

then clone nvdiffrec repository (this will be stored wherever you are cd into - in my case c:\Users\Name

git clone https://github.com/NVlabs/nvdiffrec

move nvdiffrec folder into anaconda3\envs\torch (if you aren't already cd'd into this location)

cd into nvdiffrec folder then start training an example

python train.py --config configs/bob.json --display-interval 20

---

Did it work for you?

[isaacperez]

This worked for me

[yuanzhi-zhu]

Hi, I just made it work to install tiny-cuda-nn follow the instruction here: https://docs.nerf.studio/en/latest/quickstart/installation.html

[2461011611]

嗨，我刚刚按照以下说明安装 tiny-cuda-nn https://docs.nerf.studio/en/latest/quickstart/installation.html

Hi, I just made it work to install tiny-cuda-nn follow the instruction here: https://docs.nerf.studio/en/latest/quickstart/installation.html