Has anyone got Yue working in a Ui version Windows or WSL for 5090 GPU?

[adamreading]

I have tried all day to get YuE installed on WSL - and just going round in error messages and dependency loops - it's ended up killing my whole WSL install and I am having to start again - (comfyui all gone)

Has anyone made a fully working Docker for YuE for the 5090 - with one of the front end Gradio add ons - for either WSL or Windows 11 Pro?

[a43992899]

wow 5090

How much is it right now? Maybe you can try skipping flash attention and see if it works. It could be flash attention not supporting 5090.

[adamreading]

I ordered a machine on day one, from a supplier that only builds whole systems - they had plenty of stock. I got the MSI with the liquid cooling and overclocking. In the UK, they were advertising that at 2680 UK Pounds, mine was 2340 as part of the build, ,but the whole build was 5700 UK pounds... I just realised the US Keyboard they gave me doesn't have a pound sign visible - I guess there's some code to get it,.  On Monday 10 February 2025 at 13:35:40 GMT, Ruibin Yuan ***@***.***> wrote: wow 5090 How much is it right now? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[WrongProtocol]

in response to adamreading "Has anyone made a fully working Docker for YuE for the 5090 - with one of the front end Gradio add ons - for either WSL or Windows 11 Pro?"

i run a docker on windows 11 with WSL2 running Ubuntu, nvidia-smi installed in WSL. I run a 4090.
The docker image is a fork of a fork that has the exllamav2 version of YuE and the gradio interface, which I've modified for my own needs. but it does all run. https://github.com/WrongProtocol/YuE-exllamav2-UI

note it's not meant to be a turn-key launch for anyone who downloads it, but its pretty darn close. you might have to set up the 2 voumes in docker which takes 2 seconds.

[adamreading]

in response to adamreading "Has anyone made a fully working Docker for YuE for the 5090 - with one of the front end Gradio add ons - for either WSL or Windows 11 Pro?"

i run a docker on windows 11 with WSL2 running Ubuntu, nvidia-smi installed in WSL. I run a 4090. The docker image is a fork of a fork that has the exllamav2 version of YuE and the gradio interface, which I've modified for my own needs. but it does all run. https://github.com/WrongProtocol/YuE-exllamav2-UI

note it's not meant to be a turn-key launch for anyone who downloads it, but its pretty darn close. you might have to set up the 2 voumes in docker which takes 2 seconds.

I spent about 8 hours trying to load it in wsl, in the end it broke my entire wsl load and my comfyui instance all dead and deleted. Because the 5090 has to run in cuda12.8 sm_120 torch 2.6 /2.7 and just so many dependancy fails. I’ve given up til they sort PyTorch etc out and someone starts releasing 5090 dockers /pinokio etc

[WrongProtocol]

it does seem the 5090 has some extra dependencies that are well beyond the scope of this project. Im sorry I'm not more familiar with it to help further.

[alisson-anjos]

in response to adamreading "Has anyone made a fully working Docker for YuE for the 5090 - with one of the front end Gradio add ons - for either WSL or Windows 11 Pro?"

i run a docker on windows 11 with WSL2 running Ubuntu, nvidia-smi installed in WSL. I run a 4090. The docker image is a fork of a fork that has the exllamav2 version of YuE and the gradio interface, which I've modified for my own needs. but it does all run. https://github.com/WrongProtocol/YuE-exllamav2-UI

note it's not meant to be a turn-key launch for anyone who downloads it, but its pretty darn close. you might have to set up the 2 voumes in docker which takes 2 seconds.

I forked exllamav2 and then you forked mine hehehehe

[alisson-anjos]

I'm currently trying to do this, I have a 5090 too and I've already made some progress, the most problematic is flashAttn because it has specific kernels.

[adamreading]

I'm currently trying to do this, I have a 5090 too and I've already made some progress, the most problematic is flashAttn because it has specific kernels.
Yeah I think that's what wrecked my install - it tool about 3 hours compiling a version of flashattention to use with sm_120 and after that everything went wrong lol... I went round the dependency circle for about 4 hours - with chatGPT giving me instructions, but I soon realised we were just repeating ourselves in 20 step loops - ended up scrapping the whole install, losing comfy and starting again in Windows Portable