I can create a Dockerfile for you?

[DrewThomasson]

I saw that you had the checkbox Support Docker was unchecked and was wondering if I could contribute,

If so I was a little bit confused of what you would want the Docker to do?

Support a command-line inference?

Or just do the same Gradio space?

I'm willing to put in some effort to give this repo a Dockerfile that will compile for Arm + x86 and x86 tagged variant's for CUDA 11 and CUDA 12.

And if you want, I can also create a github workflow for you that will automatically rebuild the Docker images each time there is a major change to the repo and push it to your dockerhub.

Just wanted to know your thoughts before I put effort into this.

And Thanks for making this amazing open source tool! :)

[NZqian]

Thanks so much for offering to help with this! Our team’s been pretty swamped with DiffRhythm 2 lately, so we haven’t had much time to work on other stuff. Docker support would be amazing, and we’d really appreciate your contribution!

[NZqian]

I think Docker would be super useful for providing a consistent environment to run the model—it could handle both command-line inference and the Gradio UI. Plus, I saw someone in another issue struggling with Arch Linux setup, so Docker could definitely help avoid those kinds of issues.

[DrewThomasson]

Sound good!

I'll see if I can get a Dockefile + compose out when I find time 👍

[DrewThomasson]

@NZqian

Looks like someone got on the boat for that before I did

But I made a major modification to the dockerfile it to allow easy alternative GPU build options

TORCH_VERSION Options = cuda12, cuda128, cuda11, rocm, xpu, cpu

PR -> #85

[orkutmuratyilmaz]

@DrewThomasson there are some necessary changes for your PR. have you tried it on ROCM?