Cuda 12?

[superlbr]

PyTorch 2.x?

[Fictionarry]

Hi, we have a test on PyTorch 2.1.1 and CUDA 12.1. Overall, it works well, although there seem to be some minor issues with numerical stability in some cases. The corresponding environment file is provided. However, we still don't have sufficient experience with such an environment and prefer to commend the PyTorch 1.1x.x and CUDA 11.x environment.

[justin4ai]

Installing ./submodules/diff-gaussian-rasterization and others keeps failing with the provided CUDA 12.1 environment. Do you also have a .txt requirement file by any chance?

[Fictionarry]

Installing ./submodules/diff-gaussian-rasterization and others keeps failing with the provided CUDA 12.1 environment. Do you also have a .txt requirement file by any chance?

Hi, I kept a list of dependencies that require installation with pip here, and so you can separately install the conda part and then the pip part where the problems occur. If you are looking for a pure pip version, you can first manually install pytorch with pip and then the dependencies below as a pip requirement file by yourself.

InsTaG/environment.yml

Lines 153 to 181 in 42f68f8

	# --------------- pip dependencies ---------------
	# ./submodules/diff-gaussian-rasterization
	# ./submodules/simple-knn
	# ./gridencoder
	# numpy
	# pillow==9.5.0
	# scipy
	# tensorboard
	# opencv-python
	# tensorboardX
	# pandas
	# tqdm
	# matplotlib
	# rich
	# packaging
	# scikit-learn
	# face_alignment
	# python_speech_features
	# numba
	# resampy
	# pyaudio
	# soundfile
	# configargparse
	# lpips
	# imageio-ffmpeg
	# librosa

[justin4ai]

Could you please let me know your gcc version? It still doesn't work for me..

[Fictionarry]

Could you please let me know your gcc version? It still doesn't work for me..

Both 11.4 and 8.4 were tested successfully. However I do not think it's a common problem here. Maybe you can open a new issue with the exact raised exceptions and maybe I can help you see what the problem is?

[justin4ai]

Hello, sorry for replying very late. I've been struggling to resolve the problem by myself by installing custom gcc version of 11.4 which matches to your verified one. Many of problems are resolved but the following pyaudio installation error occurs:

      src/pyaudio/device_api.c:9:10: fatal error: portaudio.h: No such file or directory
          9 | #include "portaudio.h"
            |          ^~~~~~~~~~~~~
      compilation terminated.
      error: command '/home/seungillee/custom_gcc_justin_11_4/gcc-install/bin/gcc' failed with exit code 1
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for pyaudio
Failed to build pyaudio
ERROR: Could not build wheels for pyaudio, which is required to install pyproject.toml-based projects

. I'm avoiding opening a new issue since this is not directly related to your project. Still I'm struggling to install the package without sudo privilege. Do you have any idea to go further by any chance?

[Fictionarry]

Hello, sorry for replying very late. I've been struggling to resolve the problem by myself by installing custom gcc version of 11.4 which matches to your verified one. Many of problems are resolved but the following pyaudio installation error occurs:

      src/pyaudio/device_api.c:9:10: fatal error: portaudio.h: No such file or directory
          9 | #include "portaudio.h"
            |          ^~~~~~~~~~~~~
      compilation terminated.
      error: command '/home/seungillee/custom_gcc_justin_11_4/gcc-install/bin/gcc' failed with exit code 1
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for pyaudio
Failed to build pyaudio
ERROR: Could not build wheels for pyaudio, which is required to install pyproject.toml-based projects

. I'm avoiding opening a new issue since this is not directly related to your project. Still I'm struggling to install the package without sudo privilege. Do you have any idea to go further by any chance?

Got it. Unavailable sudo does matter. Maybe you can try to manually get the header file from https://www.portaudio.com/docs/v19-doxydocs/portaudio_8h_source.html and install it by Uberi/speech_recognition#314 (comment). However, I'm not sure if it works but you can try it. A common solution should be to install portaudio first, but it often requires sudo permission.

Another solution is that since the project actually only uses the audio feature but not the audio itself during the training and inference, you can try to skip the processes that directly handle the audio file (typically on the audio feature extraction) in your server and extract the features at your PC, Google Colab, or somewhere else available. Therefore, pyaudio is not necessary anymore on the machine without sudo.