Support of other kinds of GPU

[XixuHu]

Hi Ben, I would like to know if there is any method that can support the training on RTX 3090/ A6000. Thanks a lot.

[TheLastBen]

Hi, Colab doesn't offer those GPUs as far as I know

[XixuHu]

Yeah, I know. I download the notebooks and modified the corresponding part to run on my local GPU to accelerate the training. What corresponding changes should I make if my local GPUs are RTX3090/A6000?

`if (gpu=='T4'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/T4/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='P100'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/P100/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='V100'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/V100/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='A100'): %cd /usr/local/lib/python3.7/diffusers/models/ !rm /usr/local/lib/python3.7/diffusers/models/attention.py wget.download('https://raw.githubusercontent.com/huggingface/diffusers/269109dbfbbdbe2800535239b881e96e1828a0ef/src/diffusers/models/attention.py') %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/A100/xformers-0.0.13.dev0-py3-none-any.whl `

[CptGabok]

If you get this working, let us know!

[TheLastBen]

you don't need that cell for local, try adapting this : https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/Local_fast_DreamBooth-Win.ipynb

[XixuHu]

Hi Ben. Thanks a lot! I will try it! :)

[nawnie]

Yeah, I know. I download the notebooks and modified the corresponding part to run on my local GPU to accelerate the training. What corresponding changes should I make if my local GPUs are RTX3090/A6000?

`if (gpu=='T4'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/T4/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='P100'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/P100/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='V100'): %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/V100/xformers-0.0.13.dev0-py3-none-any.whl

elif (gpu=='A100'): %cd /usr/local/lib/python3.7/diffusers/models/ !rm /usr/local/lib/python3.7/diffusers/models/attention.py wget.download('https://raw.githubusercontent.com/huggingface/diffusers/269109dbfbbdbe2800535239b881e96e1828a0ef/src/diffusers/models/attention.py') %pip install -q https://github.com/TheLastBen/fast-stable-diffusion/raw/main/precompiled/A100/xformers-0.0.13.dev0-py3-none-any.whl `

You will want to grab the compiler off the 111 ui bottom cell. It's gonna take about 50 mins but you should be a ble to pip install (location of whl) at that point, compile 2 put them in diffrent folders then you could rewrite the code with install directories from there.

Or be simple and write 1 line for each and hash out what you are not using

[Portareumbra]

Any way to make this work with the Tensor TPUs offered on other notebooks?

[nawnie]

Any way to make this work with the Tensor TPUs offered on other notebooks?

Last I checked there's is a flax/Jax version out but you miss out on a ll the speed features so it kinda levels out (as far as training)

[0xdevalias]

@XixuHu You may also like to check out this issue, where I was exploring getting things running on an RTX 3090 on RunPod:

• https://github.com/TheLastBen/fast-stable-diffusion/issues/80#issuecomment-1305002261