Should we install Deepspeed or not?

[runner22k]

For anyone who is trying to install deepspeed. Should we install Deepspeed or not?

While installing when we execute accelerate config we are asked to install deepspeed or not. If you select yes, we will be asked to run below pip. That is when I am getting this error.

Luckily I opened https://github.com/kohya-ss/sd-scripts

there developer mentioned answers too. Where we are asked to not install deepspeed.

Steps to reproduce the problem

1. In the VENV run accelerate config.
2. When asked to install deepspeed select YES.
3. Run pip install deepspeed

What should have happened?

deepshpeed should be installed. But we are getting CUDA error.

What platforms do you use to access UI ?

Windows

What browsers do you use to access the UI ?

Mozilla Firefox

Command Line Arguments

pip install deepspeed

Additional information, context and logs

(venv) PS D:\kohya_ss> pip install deepspeed
Collecting deepspeed
  Using cached deepspeed-0.7.7.tar.gz (712 kB)
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error

  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [8 lines of output]
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "C:\Users\jan\AppData\Local\Temp\pip-install-bh_2b2bj\deepspeed_337b9c20b5434938ad45f7a28ad99337\setup.py", line 78, in <module>
          cupy = f"cupy-cuda{''.join(map(str,installed_cuda_version()))}"
        File "C:\Users\jan\AppData\Local\Temp\pip-install-bh_2b2bj\deepspeed_337b9c20b5434938ad45f7a28ad99337\op_builder\builder.py", line 41, in installed_cuda_version
          assert cuda_home is not None, "CUDA_HOME does not exist, unable to compile CUDA op(s)"
      AssertionError: CUDA_HOME does not exist, unable to compile CUDA op(s)
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.
(venv) PS D:\kohya_ss>

[bmaltais]

No need to... but it should be hurt.

[Odawgthat]

I got this error too, what should I do???

[Odawgthat]

No ne

What does that mean

[bmaltais]

No idea. I suggest you raise this directly with kohya_ss since he create the actual training code. I only create the GUI around them to make easier for users.

[runner22k]

@Odawgthat I switched to Linux and google Colab, as we can't do much without deepspeed on 8GB or less VRAM GPUs and on Windows. It is available only on Linux, and it will help us run some trainings. But definitely it will take much much longer. I could have spent more money on GPU with 16GB VRAM, instead of buying 64GB system RAM. But its too late now. I realized my mistake lately.

I lost interest to train on my system. Maybe in the future I will upgrade to a new GPU with more VRAM and Linux as base, as linux has more work around options comparatively.